* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Non-subject Arguments in Indonesian - ORB
Udmurt grammar wikipedia , lookup
Macedonian grammar wikipedia , lookup
Ukrainian grammar wikipedia , lookup
French grammar wikipedia , lookup
Old Irish grammar wikipedia , lookup
Germanic strong verb wikipedia , lookup
Germanic weak verb wikipedia , lookup
Modern Greek grammar wikipedia , lookup
Japanese grammar wikipedia , lookup
Esperanto grammar wikipedia , lookup
Kannada grammar wikipedia , lookup
Chinese grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Navajo grammar wikipedia , lookup
Swedish grammar wikipedia , lookup
Malay grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Modern Hebrew grammar wikipedia , lookup
Polish grammar wikipedia , lookup
Sotho verbs wikipedia , lookup
English clause syntax wikipedia , lookup
Russian grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Hungarian verbs wikipedia , lookup
Italian grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Old English grammar wikipedia , lookup
Kagoshima verb conjugations wikipedia , lookup
Georgian grammar wikipedia , lookup
Icelandic grammar wikipedia , lookup
Yiddish grammar wikipedia , lookup
Spanish grammar wikipedia , lookup
Serbo-Croatian grammar wikipedia , lookup
NonNon-Subject Arguments in Indonesian Simon Musgrave Submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy October 2001 Department of Linguistics and Applied Linguistics The University of Melbourne i Abstract The grammatical function subject can be identified reliably in Bahasa Indonesia (Indonesian), but the same is not true of other clause-level nominal constituents in the language. The tests proposed in previous studies for identifying an object grammatical function turn out to be unreliable and inconsistent when the full range of data is considered. This thesis attempts to clarify the problem by examining non-subject arguments in Indonesian in the theoretical framework of Lexical-Functional Grammar. Both the properties associated with the various types of argument and the means by which they are licensed in clauses turn out to be problematic. Two argument verbs appear in a range of clause types which are related in interesting ways. I argue that it is possible to give a coherent analysis of the system as a whole, treating it as the basic transitive system of the language, when certain relationships are recognised as morphological rather than syntactic. This analysis also reveals similarities between Indonesian and more conservative Austronesian languages such as those of the Philippines. Another type of clause, sometimes referred to as the 'adversative verb' clause type shares an important property with one of the types of transitive clause, that of allowing an oblique argument to appear without a licensing preposition if it is adjoined to the verb which governs it. Another class of verbs, emotion and cognition verbs, has a non-subject argument which can be either a prepositional phrase or a bare nominal, but in this case the bare nominal is not an oblique argument, adjunction is not the relevant syntactic relation. The only possible analysis of these verbs, maintaining the assumption that the transitive system has been identified, is that they have a subject and a secondary object as their arguments. I argue that this analysis is implausible, and that the assumption regarding the transitive system must be reconsidered. When this is done, I claim that Indonesian is best analysed as having two distinct types of clausal organisation, one in which the major constitutents are subject and verb phrase, and one in which they are subject and predicate. Most verbs can head the predicate constituent in the second type of clause, with morphology playing a crucial derivational role. Evidence from nominalised clauses and from intransitive verbs supports this analysis, but crucial aspects of the analysis resist statement in the framework of Lexical-Functional Grammar, suggesting that the theory is overly-constrained in important areas. ii Declaration This is to certify that: i. the thesis comprises only my original work except where indicated in the preface, ii. due acknowledgment has been made in the text to all other material used, iii. the thesis is less than 100,000 words in length, exclusive of tables, maps, bibliographies and appendices. Simon Musgrave Date: iii Preface This thesis reports the results of research carried out between 1997 and 2001. The empirical basis of this research consists of a survey of texts carried out by me (the sources used are listed immediately before the other bibliographic references at the end of the thesis), example sentences elicited from speakers of the languages studied, and published material as acknowledged in the body of the thesis. Data on the Sasak language was collected as part of the collaborative Lombok and Sumbawa Research Project (Australian Research Council Large Grant A59803558, Principal researcher: Prof. Peter K. Austin); the other work reported here was my own independent research. iv Acknowledgments My greatest debt is to the speakers of Indonesian, and of the other languages discussed in this thesis, who generously shared their time and knowledge with me. In particular, Umar Muslim was an endlessly patient teacher of both Indonesian and Javanese, and was always willing to share his own insights as a linguist. Katarina Sukamto, Eli Riharti, and Lalu Dasmara all also were very helpful and Waruno Mahdi always answered my e-mails. Lalu Dasmara also helped me with Sasak data, as did Herman Suheri, Syahdan and Yon Mahyuni. I am grateful to Bukhari Daud and Samadi for assisting me with Acehnese data. I am also grateful to Helen McKay for allowing me access to her corpus of Indonesian text (and for employing me). My supervisor throughout this research has been Peter Austin, and he has made the experience enjoyable by his adroit balance of formal and informal guidance, his enthusiasm for linguistics, and his deep knowledge which he is always willing to share. During my probationary candidature, Lesley Stirling and Dominique Estival were members of my supervisory committee, and I am grateful to them both for helpful comments. The Department of Linguistics and Applied Linguistics at the University of Melbourne is a stimulating and friendly place to work. I am especially grateful to Nick Evans, who always has time (how?) to discuss problems, and always has a suggestion as to how to tackle them; to Rachel Nordlinger for her LFG library; to Anna Margetts for incisive comments; and to Rob Schmittat for technical help (and coffee). I was fortunate to do this research while the department was something of a centre of Austronesian research, and I thank my fellow students Katarina Sukamto, Anthony Jukes, Nick Thieberger and Yusuf Eades for their friendship and support. Beyond the department, the wider circle of Austronesian scholars were wonderfully welcoming to a novice. I am grateful to all of the following people for many different kinds of assistance: Sander Adelaar, Wayan Arka, Barry Blake, John Bowden, Peter Cole, Susanna Cumming, Novi Djenar, Mark Durie, Michael Ewing, Dan Finer, Margaret Florey, Bill Foley, David Gil, Gabriella Hermon, Nikolaus Himmelmann, Arthur Holmer, Marian Klamer, Waruno Mahdi, Bernd Nothofer, Bill Palmer, Andy Pawley, Stan Starosta, (and a special mention for Mary Ellen Jordan, Austronesianist rtd.),. The LFG community were also very generous; thank you to Avery Andrews, Kersti Borjars, Joan Bresnan, Chris Manning, Jane Simpson and Nigel Vincent. I am enormously grateful to the government of Australia, through the Department of Education, Training and Youth Affairs, for financial support in the form of an Australian Post-graduate Award which allowed me to work fulltime on this thesis. I am also grateful to the Faculty of Arts, The University of Melbourne for a travel grant which enabled me to attend the 8th International Conference on Austronesian Linguistics (Taipei 1997); to the Melbourne Scholarships Office for a Melbourne Abroad Scholarship and to the Department of Linguistics and Applied Linguistics for a travel grant which together enabled me to attend the 7th Austronesian Formal Linguistics Association meeting (Amsterdam 2000) and to visit the University of Manchester. I am also grateful for support from the Lombok and Sumbawa v Research Project (Australian Research Council Large Grant A59803558) which assisted me in attending the 2nd International LFG conference and the 4th Australian Linguistic Institute (Brisbane 1998). Portions of this work have previously been presented to the following meetings: Austronesian Formal Linguistics Association (Amsterdam 2000); Department of Linguistics Seminar, University of Manchester; the workshop on Voice in Austronesian, 2nd International LFG Conference (Brisbane 1998); the 2nd and 3rd Victorian South East Asian Languages Symposia (1998, 1999); the University of Melbourne Linguistics and Applied Linguistics Post-Graduate Conference (1999, 2000); the Victorian Austronesian Circle; and the Department of Linguistics and Applied Linguistics Austronesian Informal Seminar. Finally, I could never have completed this thesis without the love and support of my wife, Deborah - thank you for that, and for our two beautiful daughters, Anna and Olivia. vi Table of Contents Abstract.......................................................................................................... i Declaration................................................................................................... ii Preface ..........................................................................................................iii Acknowledgments ..................................................................................... iv Table of Contents ....................................................................................... vi Abbreviations ............................................................................................... x Glossing Practice ........................................................................................ xi 1 Introduction ...............................................................................................1 1.1 Basic assumptions ............................................................................... 1 1.2 Indonesian syntax................................................................................ 4 1.2.1 General issues ............................................................................................................4 1.2.2 Tests of subjecthood .................................................................................................7 1.2.3 Tests of objecthood ................................................................................................10 1.3 Lexical-Functional Grammar............................................................... 16 1.3.1 Types of representation..........................................................................................16 1.3.2 Correspondence principles ...................................................................................19 1.3.2.1 C-STRUCTURE TO F-STRUCTURE ................................................................................19 1.3.2.2 A-STRUCTURE TO F-STRUCTURE: LEXICAL MAPPING THEORY .......................................21 1.3.3 Functional categories in Indonesian syntax ........................................................22 1.4 The licensing of arguments................................................................ 30 1.5 The origin of Bahasa Indonesia and the data used .............................. 33 1.6 Overview of the remaining chapters ................................................... 35 2 Indonesian Transitive Clauses..............................................................37 2.1 The range of transitive constructions ................................................ 37 2.2 The Status of di- ............................................................................... 41 di- as pronoun...................................................................................................41 2.2.1 2.2.1.1 THE ARGUMENTS OF GUILFOYLE, HUNG AND TRAVIS ................................................41 2.2.1.2 THE PHRASE-STRUCTURE POSITION OF DI- .................................................................44 2.2.1.3 ACTOR DOUBLING.................................................................................................49 2.2.1.4 SEMANTIC ROLE ASSIGNMENT ................................................................................53 2.2.1.5 HISTORICAL EVIDENCE ..........................................................................................54 2.2.1.6 SUMMARY ............................................................................................................55 2.2.2 di- as verb prefix .....................................................................................................56 vii 2.2.2.1 POSTVERBAL ATTACHED PRONOUNS .......................................................................56 2.2.2.2 EXTRACTION AND TOPICALIZATION .........................................................................58 2.2.3 Summary..................................................................................................................62 2.3 Prefixed Verbs: meN-V and di-V clauses............................................. 62 2.3.1 The status of non-subject actors ..........................................................................62 2.3.1.1 THE GENITIVE AGENT HYPOTHESIS ............................................................................64 2.3.1.2 AGENT INCORPORATION .......................................................................................66 2.3.1.3 TERMS AND NON-TERMS ........................................................................................68 2.3.2 Lexically specified linking ................................................................................70 2.4 Pro-V clauses .................................................................................... 74 2.4.1 TOPIC and SUBJECT as distinct functions ............................................................74 2.4.2 Syntax and morphology in Pro-V clauses ...........................................................77 2.4.3 Pronoun actors and termhood ............................................................................78 2.4.3.1 ONLY PRONOUN ACTORS CAN BE NON-SUBJECT TERMS ...........................................78 2.4.3.2 PRONOUN ACTORS CANNOT BE NON-SUBJECT NON-TERMS ......................................78 2.4.3.3 COMBINING THESE TWO GENERALISATIONS .............................................................79 2.4.4 Morphology of undergoer subject clauses ........................................................80 2.4.4.1 A PREVERBAL MORPHOLOGICAL SLOT .....................................................................80 2.4.4.2 THE LIMITS OF MORPHOLOGY ................................................................................82 2.5 An analysis ....................................................................................... 87 2.5.1 Pro-V clauses...........................................................................................................87 2.5.2 Adding -nya to the picture ...................................................................................90 2.5.3 Adding mereka ......................................................................................................98 2.5.4 Bare DP actors in di-V clauses ..............................................................................99 2.6 Summary .........................................................................................102 3 ke- -an verbs.........................................................................................104 3.1 A description of the ke- -an verbs......................................................105 3.1.1 Functions of the circumfix ke-...-an.................................................................... 105 3.1.2 Subgroups of ke- -an verbs ................................................................................. 107 3.2 Analyses of clauses with ke- -an verbs...............................................112 3.2.1 The analysis of Kana (1986)................................................................................. 113 3.2.2 Problems with Kana's analysis............................................................................. 115 3.2.2.1 THE SECOND ARGUMENT ..................................................................................... 115 3.2.2.2 RESTRICTIONS ON BASE VERBS .............................................................................. 118 3.2.2.3 POSSESSOR SUBJECTS .......................................................................................... 119 3.3.3.4 SUMMARY .......................................................................................................... 119 3.3 A new analysis ..................................................................................120 3.3.1 What are the base predicates?......................................................................... 121 3.3.2 Properties of the ke- -an predicate ................................................................... 125 3.3.2.1 DERIVING THE ARGUMENT STRUCTURE ................................................................... 125 3.3.2.2 WHAT GRAMMATICAL FUNCTIONS?...................................................................... 129 3.3.2.2.1 Subject ...................................................................................................... 129 viii 3.3.2.2.2 The status of the second argument - predicative complements .............. 130 3.3.2.3 A LEXICAL ENTRY FOR KE- -AN.............................................................................. 134 3.3.2.4 THE CASE OF KETINGGALAN ................................................................................. 136 3.3.3 Obliques and adjunction .................................................................................... 138 3.4 Conclusion .......................................................................................141 4 Emotion/Cognition Predicates..........................................................143 4.1 A class of predicates .....................................................................143 4.1.1 Criterion of membership............................................................................... 143 4.1.2 Range of constructions................................................................................. 147 4.1.2.1 THE DATA........................................................................................................... 150 4.1.2.2 PREPOSITIONAL CONSTRUCTION .......................................................................... 150 4.1.2.3 BARE NOUN-PHRASE COMPLEMENTS ..................................................................... 156 4.1.2.4 APPLICATIVES AND CAUSATIVES ........................................................................... 158 4.1.2.5 TRANSITIVE VERBS ................................................................................................ 161 4.1.2.6 INTRANSITIVE USES ............................................................................................... 161 4.1.2.7 CONSTRUCTIONS WITH COMPLEMENT CLAUSES ...................................................... 162 4.1.2.8 NOMINAL USES ................................................................................................... 164 4.1.2.9 ATTRIBUTIVE USES ................................................................................................. 166 4.1.3 Lexical categorisation ......................................................................................... 168 4.1.3.1 NOMINAL USES OF EMOTION AND COGNITION WORDS .......................................... 169 4.1.3.2 ADJECTIVES AND VERBS IN INDONESIAN................................................................ 171 4.1.3.3 THE LEXICAL CATEGORISATION OF EMOTION AND COGNITION WORDS .................... 176 4.2 Emotion verbs and grammatical functions.........................................181 4.2.1 Transitive Clauses - SUBJ and OBJ..................................................................... 182 4.2.2 OBLθ ........................................................................................................................ 185 4.2.3 OBJθ ........................................................................................................................ 188 4.3 Formalising the analysis ..................................................................190 4.3.1 An LFG analysis ..................................................................................................... 190 4.3.1.1 TRANSITIVE OR TWO-PLACE UNACCUSATIVE........................................................... 190 4.3.1.2 LEXICAL FEATURES OF THE STIMULUS....................................................................... 195 4.3.2 P&P analyses - VP shells and internal arguments ............................................ 196 4.3.3 Summary................................................................................................................ 199 4.4 Theoretical and empirical problems ..................................................200 4.4.1 The status of OBJθ in LFG..................................................................................... 200 4.4.2 Empirical issues ..................................................................................................... 202 5 Two types of clause structure............................................................204 5.1 Strategies of relativization ................................................................204 5.1.1 Resumptive pronoun strategy ............................................................................ 204 5.1.2 Subject and reduced pronouns......................................................................... 205 5.1.3 Complementary distribution of relativization strategies ................................. 207 5.1.3.1 TOPIC-COMMENT STRUCTURES ............................................................................. 208 5.1.3.2 VOSKUIL 1996.................................................................................................... 210 ix 5.1.3.3 SUMMARY .......................................................................................................... 211 5.2 The distribution of -nya ....................................................................212 5.3 Potentially predicative elements and arguments ................................215 5.3.1 Semantic roles ...................................................................................................... 216 5.3.2 Grammatical functions ....................................................................................... 217 5.3.3 Nonpredicative uses of prefixed verbs ............................................................. 220 5.3.4 Overt copulas ....................................................................................................... 222 5.3.5 Intransitive verbs: the prefixes ber- and ter- ..................................................... 224 5.3.6 Historical and comparative evidence.............................................................. 226 5.3.7 Summary................................................................................................................ 230 5.4 Problems for an LFG account.............................................................231 5.4.1 Where does SUBJ come from?........................................................................... 231 5.4.2 Status of the non-subject argument.................................................................. 235 5.5 Conclusions......................................................................................238 Appendix: Indonesian and the accessibility hierarchy ..............................240 Sources of Examples ...............................................................................242 References................................................................................................242 x Abbreviations 1, 2, 3 ACC ADJ......ADJ AF APPL AS CAUS CLASS COLL COMP COP DAT DUP DV EMPH EXCL FOC FRAC FUT GEN HON IF INCL INT IRR LF LNK LOC NEG NOM NOM......NOM OBL ORD PERF PL POSS PROG PRT REAL REL SG UF US 1st, 2nd, 3rd person accusative adjective-forming circumfix actor focus applicative actor subject causative classifier collective comparative copula dative reduplication dative voice emphatic particle exclusive focus marker fraction future genitive honorific instrumental focus inclusive intensifier irrealis locative focus linker locative preposition negation nominative nominalising circumfix oblique ordinal perfective plural possessive progressive particle realis relative clause marker singular undergoer focus undergoer subject xi Glossing Practice All Indonesian examples are glossed consistently, regardless of the source and spelling is regularised to the post-1972 standard (see section 1.5) except for proper names. With the exception of the allomorphy of prefixes with a final nasal segment, which is described and exemplified in section 1.3, Indonesian morphology is quite transparent. Therefore the morpheme breaks are not indicated in examples. Where the morphological structure of a word is not relevant to the discussion, the morphemes are not separately glossed. Some morphemes are indicated as separate in the gloss line, but are not given a glossing. This occurs in two situations: i) the morpheme in question is the subject of analysis in the surrounding discussion; when the analysis has been established, a suitable gloss is provided in subsequent chapters; ii) the meaning of the morpheme is irrelevant to the discussion. Words which involve reduplication are consistently shown as being of the form base-reduplicant. This is a convenience, and should not be taken to imply any theoretical claim. Examples from languages other than Indonesian are generally glossed as in the original source. Any additional abbreviations used are then explained in a footnote. The exception to this principle is examples from Tagalog: analyses, and therefore glossing, of Philippine languages vary widely and I have regularised glosses for this language. NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 1 Introduction This thesis is an investigation into the properties of non-subject arguments in Bahasa Indonesia (Indonesian) in the framework of LexicalFunctional Grammar. How to identify subjects and what their properties are is generally agreed in studies of Indonesian (one exception to this generalisation is discussed later), but less work has been done on other clausal constituents. This comparative neglect mirrors the situation in syntactic studies more generally: a great deal of work has been devoted to establishing empirically valid and theoretically useful definitions of what it means to call a nominal a subject1 (see the papers in Li (ed) 1976, especially that by Keenan), but the same attention has not been devoted to other nominal constituents of clauses (but see Plank (ed) 1984 and Dryer 1986). The problems in Indonesian will turn out to be rather more complex than those encountered in many other languages. It will be argued in this work that, in Indonesian, even the nonsubject nominal of the most common type of clause with two nominal constituents cannot be described as a (direct) object without raising serious questions, both empirical and theoretical, and that accounting for how different nominal constituents are licensed to appear in clauses poses major challenges. My conclusion will be that, despite the success of LFG as a tool for analysing a wide range of typologically diverse language, the theory as currently formulated is unable to give an adequate account of the facts which will be described. The remainder of this chapter is organised as follows. Section 1.1 sets out some basic assumptions about nominal constituents of clauses, their properties and the classes into which they can be analysed. Section 1.2 introduces some basic information about Indonesian, and discusses tests which have previously been proposed as diagnostics for subjecthood and objecthood. Section 1.3 introduces enough of the theoretical apparatus of Lexical-Functional Grammar (LFG) for the reader to negotiate the analyses presented later. Section 1.4 presents the account which LFG gives of the licensing of nominal constituents in configurational languages, and discusses some Indonesian data which are contrary to the predictions made by that theory. Section 1.5 discusses the issue of variation in Indonesian and the implications it has for data collection. Finally, section 1.6 is an overview of the remaining chapters. 1.1 Basic assumptions Two sorts of properties of nominal constituents are of interest in syntactic analysis: coding properties and behavioural properties (Keenan 1976). Coding properties are manifest formal properties such as case-marking and linear position. Thus, in Latin, the two nominal constituents associated with a transitive verb have different case forms. In most instances, one is in Throughout, I use the terms subject and object in an intuitive sense. Where specific grammatical functions are referred to, I use the LFG grammatical function designators SUBJ, OBJ etc. 1 1 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 the case form traditionally termed nominative and is the subject, the other is in the case form termed accusative and is the object: 1. Puella videt mensam girl.NOM see.3SG table.ACC 'The girl sees the table.' The case forms fully identify the function of the nominals (in most instances), and word order is relatively free. An alternative strategy is for a nominal to have the same form in all functions, but to be functionally identified by its position. English uses this strategy: subjects precede verbs in unmarked order, and objects immediately follow them. It is possible for a language to use both strategies at once, to greater or lesser extent. Thus, besides linear order, English also has case forms for some pronouns. Coding properties may also be formally marked on another constituent. For example, verbs in Latin (and in English in the case of 3rd person singular present tense) vary in their form dependent on properties of one of the nominal arguments of the clause, the subject. Verbs in other languages, for example Yimas (Foley 1991), show such agreement with more than one nominal. Behavioural properties are not overt properties of nominals, but are revealed when clauses are manipulated in various ways. For example, in English, the subject nominal of the second of two conjoined clauses can be omitted if it is co-referent with the subject of the first clause: 2. Ellen took the car and drove home. Non-subject nominal constituents in English do not have this property. Another property of this type is quantifier float. In some languages, quantifiers which are normally part of a nominal constituent can appear elsewhere in the clause, but are still construed as part of the nominal. Where this is possible, it is typically allowed for some nominal constituents but not others, and the possible readings are therefore restricted: 3. Tagalog Sinusulat lahat ng mga bata ang mga liham all child NOM PL letter UF.write CORE PL 'The/some children write all the letters.' (NOT: 'All the children write the letters.') (Schachter 1976: ex16b) Other behavioural properties include whether or not a nominal can be head of a relative clause, whether it can be a controller of the reference of a missing argument in a dependent clause, and whether it can be omitted in a dependent clause (i.e. whether it can be controlled). There is a consensus across theories that nominal constituents of clauses can be divided into two groups on the basis of their coding and behavioural properties. The most visible of these are coding properties: the first type of nominal is a bare NP in a language without case-marking, or it carries one of the grammatical cases in a case-marking language (Blake 1994: 32-34). The second type typically require adpositions in a language without case-marking, they are marked with one of the semantic cases in a casemarking language. This division correlates largely, but not perfectly, with 2 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 another division: that between nominals which are necessary for the interpretation of the verb and are assigned semantic roles by it, and other nominals which are compatible with the semantic interpretation of the verb (or clause), but which are not required. The correlation breaks down in cases where a nominal is assigned a semantic role by the verb, but nevertheless is coded using semantic case or an adposition. In what follows, I will use the following terminology: arguments will refer to the nominals assigned semantic roles by the predicate of a clause; direct argument or term will refer to an argument that does not require a mediating mechanism such as an adposition; oblique argument will refer to an argument for which such a mediating mechanism must be employed. The word adjunct will refer to nonarguments, that is elements which are not assigned a semantic role by the predicate, and non-terms will on occasion cover both obliques and adjuncts. This terminology is illustrated in the following diagram: 4. For most theories, this division follows from certain assumptions about how nominal constituents are licensed to appear in clauses. The common view is that predicates have an argument list as a lexical property. This is modelled as argument structure in LFG (Bresnan 2001a: chapter 14), and as the theta grid in Principles and Parameters theory (Haegeman 1991: 41-46). This argument list gives the valence of the verb, the number of arguments it syntactically associates with, and interacts with the rules which govern the projection of phrase structure in the language. One result of this interaction, at least in configurational languages, is that there are some arguments whose presence is licensed just by the phrase structure, and there are other arguments whose licensing requires the mediation of some other mechanism. For example, the English verb put has an argument list consisting of two terms and an oblique. This information interacts with the phrase structure rules of the language so that one complement position is projected as sister to the verb, and this can be occupied by an NP without further ado. The NP is licensed to appear there by the lexical properties of the verb and the phrase structure. But the lexical properties of the verb, the fact that it is transitive not ditransitive, only allow one such position to be generated. Phrase structure rules allow the generation of a specifier position in the maximal projection, 3 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 and any number of adjunct positions, but these positions do not license arguments in the same way that the complement position does2. Some additional mechanism is necessary, and in English the strategy used is to code the argument as a prepositional phrase. These principles account for the difference in grammaticality between the following two sentences: 5. 6. *John put a chair the garden. John put a chair in the garden. The preposition actually used to mediate the licensing of the argument NP contributes to the interpretation of the clause. There are cases in English where the minimally contentful preposition of can be used as a pure case marker, but this is not possible with put. This insight also lies behind the distinction mentioned above between grammatical and semantic cases (Blake 1994: 32-34). LFG uses a version of the theory sketched above to account for the licensing of nominals; I will lay this out in detail in section 1.4 after the basics of Indonesian syntax have been described and the apparatus of the LFG theory has been introduced. 1.2 Indonesian syntax 1.2.1 General issues Bahasa Indonesia is the national language of the Republic of Indonesia and is spoken by about 170 million people, although it is not the first language of the majority of these speakers (Grimes ed. 1996). Further historical details are given in section 1.5, which also discusses the issue of variety in Indonesian and the implications this has for data collection. The phonology of the language is not of interest in this study; the interested reader is referred to Moeliono & Dardjowidjojo (ed. 1988) for information. Orthography is straightforward except for e which represents both schwa and a mid-high front vowel. In my examples, following standard practice, no attempt is made to distinguish these. The entries in the dictionary of Echols & Shadily (1994, afterwards E&S) provide this information. In addition, j represents a palatal glide, and c represents a voiceless affricate. ng represents a velar nasal, and if a voiced velar stop follows, ngg is written. Following Sneddon (1996), I assume that Indonesian has a standard set of lexical categories: noun, verb, adjective, preposition and adverb. However, the question of whether adjectives are a category distinct from verbs is not clearcut, and is discussed in section 4.1.3, and chapter 5 will argue that the maximal projection of verbs should not always be analysed as VP. The question of whether the functional categories I (or whatever complex of categories is assumed after Pollock (1989)) and D exist in the language is discussed in section 1.3.3. I ignore here the possibility that the subject is generated in Spec of VP (see references cited by Haegeman (1991: 327, n22). In any case, such proposals agree that the argument is not licensed in that position. 2 4 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 The morphological system of Indonesian is unusual in that it is exclusively derivational3. There is no marking for case or gender on nouns, verbs are not marked for tense, aspect or mood, nor is there agreement or cross-referencing. Many verbs can take the prefixes meN- and di-, traditionally analysed as marking active voice and passive voice respectively. The discussion of chapter 2 will show that this analysis is not correct, and chapter 5 will suggests that these prefixes are in fact derivational also. The prefix meN- (and the nominalising prefix peN-) end with a nasal segment which assimilates to a following consonant. When the consonant is unvoiced, the nasal fuses with it, except in the case of the affricate c. Before a vowel, the form of the prefix is meng-, and before a liquid the nasal segment deletes. The following examples illustrate this allomorphy: 7. Base anggap baca dapat ganggu peras tulis kasih seluruh curiga lihat Prefixed form menganggap membaca mendapat mengganggu memeras menulis mengasi menyeluruh mencuriga melihat Aspects of the verbal morphology of Indonesian have been studied in detail, in particular the suffixes -i and -kan (Arka 1993, Kana 1986, Tampubolon 1983, Voskuil 1996). These two suffixes can derive verbs from noun or adjective roots (examples 8 and 9), and in some cases can derive verbs from other verbs without changing the valence, but with a different meaning (examples 10 and 11). Both suffixes can increase the valence of a verb as applicatives, that is they add a non-subject to the argument array (examples 12 and 13). The result can be a verb with three arguments, none of them introduced by prepositions. -kan is also a causative, in which case it adds an agent subject to the argument array4. Where the base is not a verb, or is a one-argument verb, the result is a two argument verb (example 9). Where the base is a two argument verb, the derived causative verb has two direct arguments with the original actor appearing as an oblique argument (example 14): The sole exception to this generalisation is the use of reduplication. This morphological process can indicate plurality when applied to nouns and continued or iterated action when applied to verbs. However, in each case other meanings are possible also (Sneddon 1996: 1521) 3 There are occasional instances where -i has causative effect. See section 4.1.2.4 for an example. 4 5 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 8. Pintu dan jendela diwarnainya door and window di.colour(n.).APPL.3 'He painted the doors and windows green.' hijau green 9. Ia menerangkan 3SG meN.clear(adj.).CAUS 'He explained his situation.' 10. Ia mendapat dia di 3SG meN.meet 3SG LOC 'He met her at her house.' 11. Polisi mendapati bengkel pemalsu police meN.meet.APPL workshop forger 'Police found the counterfeiter's workshop.' 12. Dia mengirimi suaminya 3SG meN.send.APPL husband.3 'She sent her husband money.' 13. Dia menuliskan ayahnya 3SG meN.write.APPL father.3 'She wrote a letter for her father.' 14. Saya memeriksakan mata ke 1SG meN.examine.CAUS eye to 'I had my eyes examined by the doctor.' keadaannya situation.3 rumahnya house.3 uang money uang money surat letter doktor doctor Indonesian morphology as a whole has not yet been analysed satisfactorily. In particular, there are restrictions on the co-occurrence of affixes which are not well understood. One example of this is discussed in section 3.3.1. The lack of inflectional morphology means that the major coding property available to indicate the function of terms in Indonesian is linear order. The unmarked word order in a clause is for the subject to precede the predicate, and for non-subject arguments to follow the head of the predicate. Given these facts and the lack of other coding properties, I will assume that Indonesian is a configurational language without arguing the position in detail. That is, I assume that there is a subject position which is outside the phrase-structure constituent which includes the positions occupied by nonsubject arguments. Pronouns provide the only exceptions to the generalisation given above regarding coding properties. The third person pronoun has two forms, remnants of a more extensive case-marking system, and one of these has limited distribution which is discussed in section 1.2.2 below. In one clause type, proclitic forms of first and second person pronouns occur, but I argue in chapter 2 that this is a phonological phenomenon. Finally, enclitic pronouns are possible in some positions. The distribution of these forms is important in the argumentation of each chapter in this work, and in chapter 5, I suggest 6 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 that their distribution is one of the crucial diagnostics revealing that the Indonesian has two types of clausal structure existing side-by-side. Coding properties of arguments can offer only limited evidence in the investigation of the status of nominal constituents in Indonesian. Most previous studies have therefore relied more on evidence from behavioural properties. The following sections review some of this work. 1.2.2 Tests of subjecthood Traditional grammars (McDonald & Darjowidjojo 1967, Sneddon 1996 [afterwards IRG]) and generative studies (Chung 1976a,b, Kana 1986, Arka & Manning, to appear) have agreed that identifying the subject of an Indonesian clause is a straightforward matter, with one clause type being an exception. All agree that the basic word order is subject-predicate, with all major categories possible as heads of the predicate. V is initial in its projection, therefore the word order of verbal clauses is apparently S - V - (O), and in most cases the first noun phrase in a clause can be taken to be subject. As seen in examples 12 and 13 above, two non-oblique arguments can follow the verb. In addition to the derived three argument verbs in the examples above, there are a few basic verbs which take three direct arguments such as beri 'give' and ajar 'teach'. There are additional tests available which confirm the word order generalisation for subjects, and enable doubtful cases to be decided. Firstly, the third person singular pronoun has two forms, dia and ia. The first of these can occur in any environment, but the second is restricted to three possibilities. It can appear whenever it is subject of the clause; it can appear as the agent in a marked clause type discussed in more detail below; and it can appear in the position of object when it is also the subject of a complement clause: 15. Dia/ia akan menolong meN.help 3SG FUT 'She will help us.' 16. Kami akan 1PL.EXCL FUT 'Us she will help.' 17. Kami akan menolong 1PL.EXCL FUT meN.help 'We will help her.' 18. Saya menganggap meN.consider 1SG 'I consider him stupid.' dia/ia 3SG kami 1PL.EXCL tolong help dia/*ia 3SG dia/ia 3SG bodoh stupid It is a characteristic of the clause type exemplified in 16 that the agent pronoun must be adjacent to the verb and the positioning of auxiliaries establishes that a clause is of this type. Such structures are thus easily 7 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 identified, and in all other cases, the possibility of ia occurring indicates that it is in a subject position. Example 18 introduces the next test which can be used: in structures such as example 18, with certain matrix verbs, the subject of the embedded clause can be treated as object of the matrix clause and can become subject of a passive: 19. Ali menganggap soal itu Ali meN.consider problem that 'Ali believes the problem is settled.' beres settled 20. Soal itu dianggap Ali beres problem that di.consider Ali settled 'The problem is believed by Ali to be settled.' This is not possible for any other nominal constituent of the embedded clause: 21. Mereka menugasi saya melakukan meN.assign 1SG meN.carry.out 3PL 'They assigned me to carry out observations.' observasi observation 22. *Observasi ditugasi mereka saya melakukan observation di.assign 3PL 1SG meN.carry.out (FOR: Observations were assigned by them for me to carry out.') Therefore, in clauses of this type, the coding properties of the post-verbal pronoun are those of a subject (ia is possible), but its behavioural properties are those of a non-subject argument. The other test which identifies subjects reliably is that it is only the subject position which can be omitted from an embedded clause and have its reference controlled by a constituent in a higher clause. This test is often exemplified with volitive verbs (e.g. Kana 1986: 27), but some of this class of verbs can also be auxiliaries in Indonesian (see discussion in section 4.1.2.7), and to avoid this complication I use examples of control into purpose clauses. Such clauses are introduced by the preposition untuk 'for', the subject must be co-referential with some nominal in the matrix clause5 and it is obligatorily absent: 23. Dia minta uang untuk (*dia) pergi ke Indonesia 3SG ask money for go to Indonesia 'He asked for money to go to Indonesia.' (Kana 1986: 28, ex.30) Even if there is a coreference relation between a nominal of the matrix clause and a non-subject in the embedded clause, the nominal must be present in the embedded clause: 5 There is some discussion of possible controllers in the following section. 8 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 24. *Saya mencari Ali untuk saya membunuh 1SG meN.look.for Ali for 1SG meN.kill (FOR: 'I looked for Ali in order to kill (him).' ) (Kana 1986: 28 ex.31) 25. *Saya mencari Ali untuk membunuh 1SG meN.look.for Ali for meN.kill (FOR: 'I looked for Ali in order to kill (him).' ) (Kana 1986: 28 ex.32) 26. Saya mencari Ali untuk 1SG meN.look.for Ali for 'I looked for Ali in order to kill him.' membunuhnya meN.kill.3 These tests make it possible to reliably identify subjects in Indonesian. Kana (1986) also claims that relativization is a test for subjecthood. This is discussed in the following section, as Chung (1976a) claims that it is a test for both subjecthood and objecthood. It was mentioned previously that there was one case in which general agreement as to the identification of subject was not complete. This is the case of the clause type seen in 16 above, which has the patient as left-most nominal, and the agent immediately to the left of the verb. The evidence of coding properties is thus ambiguous, and as example 16 shows the pronoun ia can appear as agent in such a clause. In a detailed examination of such clauses, Chung (1976b) demonstrates beyond doubt that it is the patient nominal which has the behavioural properties associated with subject in this clause type. The ia test is not relevant, but the other two tests discussed above provide clear evidence for the analysis: 27. Mereka anggap buku ini sudah saya baca 3PL consider book this PERF 1SG read 'They believe this book, I have read.' (Chung 1976b: ex 19) 28. Buku ini dianggap mereka sudah saya baca 1SG read book this di.consider 3PL PERF This book is believed by them to have been read by me.' (Chung 1976b: ex 20) 29. Mereka membeli ikan itu supaya ikan 3PL meN.buy fish that so.that fish saya masak 1SG cook 'They bought the fish so that the fish, I could cook.' (Chung 1976b: ex 28a) 30. Mereka membeli ikan itu untuk dapat saya masak 3PL meN.buy fish that for can 1SG cook 'They bought the fish to be cooked by me.' (Chung 1976b: ex29a) itu that dapat can Rather surprisingly in view of such evidence, later work occasionally takes the question to be still open (e.g. Cartier 1979). Alsagoff (1991, 1992) presents evidence from one dialect of Bahasa Melayu, the national language of Malaysia 9 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 which is almost identical to Bahasa Indonesia, that the control test (example 30) is not reliable in this case. This data is discussed in section 2.4.1. 1.2.3 Tests of objecthood There is far less consensus in the literature as to the status of various non-subject arguments in Indonesian. Chung (1976a) and Kana (1986) both give accounts of what they take to be properties of (direct) objects in Indonesian, but the two accounts differ considerably. Neither is accurate in my opinion, and neither can account for the full range of data examined in this work. Both agree that the object normally immediately follows the verb, although both allow that other possibilities exist: Kana (1986: 28) allows an adverb but not a major clausal constituent to intervene, while Chung (1976a) allows that the postverbal position 'may be changed by scrambling' (42). When a wider range of clause types is considered, positional evidence is of little value. There are several clause types in which a nominal follows the verb, but in each case expression as a prepositional phrase (PP) is also possible. In two of these cases, the nominal is obligatorily adjacent to the verb unless the PP option is chosen: 31. Buku book itu that 32. Buku itu dibaca tampaknya book that di.read apparently 'The book was apparently read by Ali.' 33. Kantor office 34. Kantor itu kelihatan dengan mudah office that ke.see.an with easy 'That office is easily visible to the police.' itu that dibaca di.read (*tampaknya) apparently kelihatan ke.see.an (*dengan with Ali Ali oleh by Ali Ali mudah) easy polisi police oleh by polisi police In the third case, the adjacency requirement is similar to that proposed for object by Kana, but expression as a PP is also possible: 35. Saya 1SG takut afraid sekali very lelaki man itu that 36. Saya takut sekali dengan lelaki 1SG afraid very with man 'I am very much afraid of that man.' itu that It will be shown below, using other tests, that the postverbal nominals in examples 31 and 33 are obliques which are realised in an unusual way, and that the postverbal nominal of example 35 is an object of some sort, but not a direct object in the sense intended by either Chung or Kana. 10 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 Both these scholars also take the possibility of being made subject in a passive clause to be a test for objecthood6. This test correctly rules out the postverbal arguments of examples 31 and 33: 31 is already a passive, and 33 is intransitive. But example 35 is still a problem. There is no passive form of this clause, unless the verb is affixed with the applicative suffix -i: 37. *Lelaki itu ditakut saya 38. Lelaki itu ditakuti man that di.afraid.APPL 'That man is feared by me.' saya 1SG Therefore this test also is not reliable. It has some value, in that in double object structures only the first object can be made subject: 39. Ali megirimi wanita Ali meN.send.APPL woman 'Ali sent the woman a letter.' itu that sebuah CLASS 40. Wanita itu dikirimi sebuah woman that di.send.APPL CLASS 'The woman was sent a letter by Ali.' surat letter 41. *Surat itu dikirimi wanita itu letter that di.send.APPL woman that (FOR: 'The letter was sent the woman by Ali.') surat letter oleh by oleh by Ali Ali Ali Ali Even this evidence is essentially redundant, as linear order is rigid in such constructions. Chung claims that three other tests are useful in identifying objects: relative clause formation, reflexive possibilities, and controlling into purpose clauses. Kana argues that: 1) Chung misinterprets the data on relative clause (1986: 25), a conclusion with which I agree, but I will argue that Kana misinterprets certain relative clause data also; 2) Chung's data on reflexivization are wrong but that the test is still useful (1986: 29-31); and 3) does not discuss the third property. I will discuss these in reverse order, as the issues raised increase in complexity going in that direction. The subject property of being potentially a controllee was discussed above with examples from purpose adjuncts. Chung (1976b) claims that deletion of the subject of a purpose adjunct clause is obligatory 'when the Chung (1976a) takes the possibility of forming a clause of the type of example 18 above as a separate test for objecthood. It is a very reliable generalisation that if the passive with di- is possible, then so is this construction: see discussion in chapter 2. 6 11 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 subject of such a complement is coreferential with some NP in the higher clause' (66), but Chung (1976a) claims that deletion is only possible when 'the subject of such a complement is coreferential with the subject or direct object of the higher clause' (48). This latter claim is directly contradicted by evidence from underived ditransitive verbs7: 42. Saya memberi Ali buku itu untuk meN.give Ali book that for 1SG 'I gave Ali the book to read.' (LIT: 'I gave Ali the book for to be read by him.') dibacanya di,read.3 43. *Saya memberi Ali buku itu untuk 1SG meN.give Ali book that for (FOR: 'I gave Ali the book for him to read it.') membacanya meN,read.3 The verb beri is an underived ditransitive, and the recipient is the direct object by linear position and would be subject of a di- passive clause. But it is the second object which can control the reference of the omitted subject in a purpose clause, and not the direct object. Therefore, although there is clearly an interesting problem to be investigated here, this test cannot be used as a reliable diagnostic for objecthood. The basis of all reflexive pronouns in Indonesian is diri 'self'. This morpheme can occur alone, or with the addition of a possessive pronoun: diri saya / diriku 'myself', or with the possessive and the emphatic reflexive sendiri: diriku sendiri 'myself'. Chung (1976a) and Kana (1986: 29-31) both claim that the possibilities for reflexives in object position are restricted in some way, but they disagree as to the facts. Chung suggests that diri plus possessive pronoun can occur in object position, but not elsewhere, while Kana suggests that diri with optional possessive pronoun is obligatory in object position when the object is coreferential with the subject, but is optional elsewhere. The crucial example is the following: 44. Kakak saya membeli sepasang sepatu meN.buy one.pair shoe older.sibling 1SG untuk dirinya for self.3 'My older sister bought a pair of red shoes for herself.' (Chung 1976a: ex11b, Kana 1986: 31, ex45) merah red For Chung, this sentence is ungrammatical as it stands and acceptable to some speakers with the addition of sendiri, while for Kana it is fully acceptable as given. Neither author remarks on the possibility for reflexives such as dirinya to appear in subject position, discussed by Arka & Manning (to appear) and in section 2.3.1.3: Chung (1976a) also claims that (in Relational Grammar terminology) only initial 1s and 2s can be controllers in this construction, therefore derived ditransitive verbs are not relevant to the discussion. 7 12 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 45. Dibandingkannya dirinya dengan raja di.compare.3 self.3 with king 'He compared himself with the king.' (LIT: 'Himself be compared by him with the king.' (IRG: 256) This is also possible for the subject of an embedded clause following a complementizer: 46. Mereka percaya bahwa dirinya lebih unggul 3PL believe that self.3 more superior orang lain person other 'They believe that they are superior to other people.' (IRG:206) dari from The distribution of the various reflexives is thus not a reliable guide to the status of arguments. In addition, my own experience in elicitation is that even the contrasts discussed here are hard to reproduce. In particular, speakers exploit the pragmatic effect of emphasis with sendiri frequently and find it hard to separate pragmatic acceptability and grammaticality. Therefore, I conclude that this test also is of limited use in identifying objects, and I will not use it in this study. The greatest difference between the approaches of Chung and Kana is seen in the question of relativization. The major strategy for forming relative clauses in Indonesian is to omit the head nominal from the embedded clause which is introduced by the invariant particle yang: 47. wanita yang pulang itu go.home that woman REL 'the woman who went home' The question at issue is what grammatical function can be borne by the gapped nominal in the embedded clause. Chung (1976a) claims that both subjects and direct objects can be gapped, while Kana (1986: 24) follows a traditional line in claiming that only subjects may be relativized in this way8. Kana argues in detail that Chung has misinterpreted her data, and is correct in this, but her position is also mistaken. Chung bases her claim that objects can be relativized on examples such as the following: 48. Ikan yang saya masak untuk Ali tidak fish 1SG cook for Ali NEG REL rasanya taste.3 'The fish that I cooked for Ali didn't taste good.' (Chung 1976a: ex35a) enak good As Kana rightly points out (1986: 24-25) an example like this is ambiguous between being an example of the construction seen in example 16 and a 8 There is also a resumptive pronoun strategy - see chapter 5 for detailed discussion. 13 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 regular transitive clause. In the first case, the patient is the subject as discussed in section 1.2.2, and there is no question of an object being relativized. Kana concludes from this that only subjects can be relativized with the gapping strategy, but she does not address a further point raised by Chung. Chung noted that the clause type seen in 16 is restricted to pronoun agents, but that examples parallel to 48 occur with other NPs as agent: 49. Kau melihat ikan yang anak itu meN.see fish child that 2SG REL 'Have you seen the fish that the child cooked?' (Chung 1976a: ex38a) masak? cook Both the cited scholars seem to be aware of the data which uncover the true generalisation, but neither makes the necessary step. Thus, Chung comments that: meng- can appear on the verb of the relative clause if the relativized noun is a subject, but it does not appear if the relativized noun is a direct object (51) and Kana notes that masak is a verb that normally appears without a prefix9 (1986: 25). The quoted statement from Chung should be extended to include the case of di- prefixed verbs: although non-subjects can never be relativized in this case, the prefix must appear when the subject is relativized. The correct statement of the constraint on gapped relative clauses therefore is that a non-subject argument can be gapped in the absence of a verb prefix. This is true for ordinary transitive verbs, all of which can be used without a prefix in a clause type considered by prescriptive grammarians to be substandard, although it is widely used. It is also true of the type of clause exemplified above in 35, as pointed out by Stevens (1970): 50. orang yang anak saya person REL child 1SG 'the person that my child saw' 51. lelaki yang saya man 1SG REL 'the man that I fear' takut afraid lihat see itu that itu that There is no difference in propositional meaning between the two clauses that follow: 52. 53. Anak saya lihat orang itu child 1SG see person that Anak saya melihat orang itu child 1SG meN.see person that 'My child saw that person.' The normal account is that this verb and a handful of others such as minta 'ask', minum 'drink' and mohon 'ask' have a fossilised prefix included in the root - see section 2.1. 9 14 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 but the non-subject argument can head a gapped relative clause based on example 52 but not one based on example 53. Two interpretations of this data suggest themselves: either the non-subject argument in both of the above clauses is an object, and relativization is not a relevant test for objecthood; or it is a relevant test and the non-subject argument of one clause should be analysed as an object but not the other. Chapter 5 examines this question in detail. Kana discusses one additional test which she claims is relevant to the identification of objects, one which Chung does not mention, and which is also very germane to the issues raised in the previous paragraph. Reduced forms of the singular pronouns exist10 which never occur independently, but which attach to a preceding preposition when the pronoun is its complement, to a preceding noun (or some nominal projections) when the pronoun is a possessor, and to some verbs when the pronoun is the non-subject argument. This last possibility is the one of interest here: Kana claims that 'pronominal direct objects may have the full form or the enclitic form' (1986: 29), and that this property is therefore a diagnostic for objecthood. But the facts here closely parallel those relating to relativization, again it is the presence or absence of a verb prefix which is crucial: 54. 55. *Anak saya lihatnya see.3 child 1SG Anak saya melihatnya child 1SG meN.see.3 'My child saw him/her' Clauses of the type seen in example 35 do not allow enclitics either: 56. *Saya takutnya 1SG afraid.3 (FOR: 'I am afraid of him.') Thus this data raises exactly the same question as the relativization data and will be discussed in detail in chapter 5. This survey shows that none of the tests proposed by Chung (1976a) or Kana (1986), including those proposed by both, is an unequivocal test for objecthood in Indonesian. Rather, close examination of the data shows that some of the tests raise questions about what can legitimately be described as an object in the language. In the present study, I will rarely discuss the properties of subjects, but where this is necessary I will use the tests discussed in section 1.2.2 which I accept as reliable. As far as non-subject arguments are concerned, I will rely on quantifier float as a test which discriminates between direct arguments, or terms, and all non-terms (this test is introduced and discussed in chapter 2). How the distribution of the various properties discussed above correlates with In the case of the 3rd person pronoun, the reduced form can have plural reference for some speakers. See chapter 2 for discussion. 10 15 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 the status of various non-subject arguments will be a major focus of the remainder of this study. 1.3 Lexical-Functional Grammar 1.3.1 Types of representation Lexical-Functional Grammar (LFG) is a theory of grammar developed in the late 1970s by Joan Bresnan and Ronald Kaplan. The theory models language as a set of parallel representations, each of which has to satisfy the constraints appropriate to that type of representation, and which together have to satisfy constraints on the correspondence between types of representation. Foundational material articulating the theory is presented in Kaplan and Bresnan (1982) and this version will be referred to in what follows as classical LFG. More recent developments are set out in Bresnan (2001a) and the bibliography of this work provides an up-to-date guide to work within the LFG framework. A brief introduction to the theory is Austin (to appear), and Falk (to appear) is an introductory text. The advantage of using LFG as a framework for syntactic investigation is that it provides a rigorous formalism, the mathematical foundations of which are well worked out, which is also flexible enough to provide useful and insightful descriptions of languages with widely varying typological characteristics. The current work aims to take advantage of the rigour made available by LFG rather than introducing new theoretical developments into the framework, although chapter 5 will suggest some limitations in the theory as currently formulated. This rigour, it is hoped, will clarify the issues to be tackled here and constrain the possible solutions which can be advanced in a way that leads to interesting hypotheses. A wide range of languages have been studied using LFG, therefore a second advantage of using the framework is that the analyses proposed will be stated in a form that allows of crosslinguistic comparison in a simple fashion. The name of the theory reveals two areas of grammar which are accorded especial importance in the theory. Firstly, a great deal of information is taken to be best treated as lexical information, rather than as part of syntax. One manifestation of this principle is that the principle of lexical integrity is basic: morphological processes are part of the lexicon and the internal structure of words plays no part in syntax. Secondly, grammatical functions (or grammatical relations) are treated as generalisations which capture equivalent mappings between predicate-argument relations and formal expression. These functions are not primitives, as the theory assumes that they can be decomposed (see section 1.3.2.2 below), but they are not reducible to phrase structure configurations or any other specific formal expression. This position is based on the observation that cross-linguistically it is relatively easy to state generalisations in terms of grammatical functions and relatively hard to state the same generalisations in terms of constituent structure or other formal characteristics. LFG models grammar as a series of parallel representations with independent properties which map onto one another in a principled fashion. Three types of representation will be of interest in the current study. Firstly, 16 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 there is the representation of constituent structure, or c-structure. Cstructures represent the constituency facts of the surface strings of languages using a standard version of X-bar theory11. The resulting representations are however rather different from those used by the Principles and Parameters theory (P&P) and its relatives. Empty categories are not a part of LFG cstructures: lexical integrity ensures that only morphologically complete words can be the terminal nodes in c-structure trees, and a principle of economy of expression (Bresnan 2001a: 91) bars the generation of nodes (except terminals and preterminals) which are not required by independent principles such as completeness, coherence and semantic expressivity (see discussion of these principles below). Functional heads and their projections are possible cstructure elements. The following is a partial c-structure for the English sentence The man will eat his dinner: 57. In addition to having all nodes fully expanded, c-structure representations in LFG normally also include functional annotations on each node. These are not part of the c-structure; they specify a part of the correspondence between the c-structure and the associated functional structures. These annotations will be further discussed and exemplified below. Functional structure or f-structure is the representation which includes information about functional relations. Functional here has two senses: this is the representation in which grammatical functions (GFs) are explicitly represented, and the way in which these are modelled can be read as function application in the mathematical sense. These representations take the form of attribute–value matrices (AVMs). An attribute can take an atomic value, or its value can be a complex structure, another AVM, but each attribute must have a unique value12. Each lexical head appears in f-structure as the value of a Trees which do not conform to X-bar theory are allowed in the description of nonconfigurational languages. The flat structure of such languages is modelled by phrase structure rules of the type S → X* , that is, a sentence consists of any number of elements in any order. Unfilled head positions are also possible in some circumstances, see discussion in section 1.3.3. 11 12 The contrary is however not the case: different attributes may have the same value. 17 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 PRED attribute, along with a list of the GFs which it assigns (if any)13. Aside from optional elements (adjuncts, for example), only grammatical functions licensed by the lexical head of an f-structure can appear in it. This requirement is referred to as the coherence condition: all GFs must be required by a predicate in the local f-structure. Also, each GF in the f-structure must have a value. For some types of dependency, this value may be obtained from outside the local f-structure by structure-sharing, but the GF attribute nevertheless has a value which can be computed within the local f-structure. This condition is referred to as the completeness condition: an f-structure must contain all the GFs which the local predicate requires. In addition to PRED attributes and GF attributes, f-structures also include morphosyntactic features such as tense or number as attributes. The following is the f-structure associated with the English sentence used as an example previously14: 58. A structure such as this can be interpreted as a collection of mathematical functions in the following fashion. Each complete f-structure is bracketed, and each can be read as a function which assigns values to the attributes it includes. Thus if we call the outermost f-structure in 58 f1, we can describe it as a function which returns certain values for various attributes: 59. f1 (PRED) = 'eat' f1 (TENSE) = fut f1 (SUBJ) = f2 f1 (OBJ) = f3 The process could be repeated for the f-structures modelling the SUBJ (f2) and the OBJ (f3). LFG assumes that f-structure is the linguistic representation which is most similar cross-linguistically. In other words, the f-structure 58 is assumed to be identical or at least very similar to the f-structure associated with the translation equivalent of the English sentence The man will eat his The value of the PRED attribute is assumed to be some sort of semantic representation of the lexical head. Following common practice, I use the orthographic form of the word in inverted commas as a promissory note for this fuller representation. 13 The POSS(essor) function is optionally assigned by most English nouns via a lexical redundancy rule. 14 18 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 dinner in other languages, regardless of how widely the c-structure representations might vary across languages. The third type of representation which will be discussed in this work is argument structure, a-structure. This type of representation models the relations between predicates and their arguments. The semantic entry associated with a predicate has a list of the semantic roles assigned by the predicate, but no information about how they are realised. F-structures have information about the GFs assigned by a predicate, but if this information is not part of the predicate's lexical entry (see next section for discussion), some additional representation is necessary which contains the minimum syntactic information necessary for the linking of semantic roles to GFs to be accomplished. A-structure is the additional representation required, and it combines information about the syntactic and semantic prominence of arguments. Terms are more prominent than non-terms, and the a-structure respects this division. Within each division, arguments are ordered by semantic prominence as reflected in the thematic hierarchy (Gruber 1965). The a-structure representation of the sentence used as an example previously would be: 60. eat < x,y> For comparison, the structure associated with the English verb put, which assigns an obligatory oblique GF, would be: 61. put < x,y:z > where the colon indicates the division between terms and non-terms. Argumentation motivating this type of representation is presented in Manning (1996a) within the framework of Head-Driven Phrase Structure Grammar, but the arguments apply equally to LFG. Other work which makes crucial use of the properties of a-structure includes Arka & Manning (to appear) and Wechsler and Arka (1998). A different version of a-structure is presented by Bresnan (2001a: chapter 14); in this style of structure the division between terms and obliques is not an explicit part of the representation and is instead derived by the effects of the Lexical Mapping Theory (see section 1.3.2.2). The question of which type of representation should be preferred is discussed in sections 2.5.2 and 3.3.2.3. The latter section discusses a case in which the two theories make different predictions, and suggests that the more articulated a-structure is supported by the data in question. In the discussion of LMT below (section 1.3.2.2), I use Bresnan-style representations to reflect the sources accurately. 1.3.2 Correspondence principles The three parallel representations discussed above, as well as others such as semantic structure which are not discussed here, are mapped onto each other by correspondence principles. The ones of interest here are those that map from c-structures to f-structures, and those that map from astructures to f-structures. 19 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 1.3.2.1 C-STRUCTURE TO F-STRUCTURE Every node in a c-structure must correspond to some f-structure, and this is modelled by annotations on c-structure nodes. These annotations may be introduced by lexical entries or phrase structure rules in special cases, but in general they are provided by default principles (Bresnan 2001a: 102-103). The annotations take the form of equations such as ↑ = ↓, where ↑ can be read as 'the f-structure of my mother node' and ↓ as 'my f-structure'. The equation given in the previous sentence is therefore interpreted as saying 'the fstructure of my mother node is identical to my f-structure'. The annotated version of the tree in 57 is the following: 62. The equation on the left-most DP node means 'the f-structure of the SUBJ in the f-structure of my mother node is identical to my f-structure'. In this fashion, information from lexical entries is combined into complex f-structures. The process can be given a rigorous mathematical treatment, which will not be set out here (see Bresnan 2001a: 56-60 for details), except to note that if the functional description of a c-structure can be resolved, then there is an infinite set of f-structures which will be compatible with the c-structure, all of which have additional attribute value pairs not mentioned in the functional description. The minimal f-structure which is compatible with the description is the linguistically interesting one15. For languages in which phrasal categories have endocentric organization, the functional annotations of c-structure nodes are predictable The account of how f-structures are built up presented here is the standard one. Recent work by Andrews and Manning (1999) has suggested that the use of complete equation as the main tool is misleading. They suggest that more insightful analyses can be given in some cases if only part of the functional information associated with some nodes is passed up to the mother node, rather than assuming total identity of the f-structures. This work does not bear on the issues considered in the present study. 15 20 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 (Bresnan 2001a: 98-109). I take Indonesian to be a language of this type, and I will therefore omit functional annotations from trees in most cases. 1.3.2.2 A-STRUCTURE TO F-STRUCTURE: LEXICAL MAPPING THEORY The prototypical association of certain semantic roles with syntactic functions, such as that holding between the agent role and the subject function, was already a part of the Sanskrit grammatical tradition (Blake 1994: 65-67). Fillmore (1968) made such associations a part of generative grammar, and much subsequent work has been devoted to analysing which facts about the semantics of verbs are predictors of the linking between semantic roles and syntactic functions (e.g. Carter 1976, Dowty 1991, Foley & Van Valin 1984, Grimshaw 1990, Jackendoff 1990, Tenny 1994). In classical LFG as described in Kaplan and Bresnan (1982), the lexical entries of verbs directly specified the GFs associated with their arguments, but subsequent work has tried to capture linking regularities through a separate module of grammar known as Lexical Mapping Theory (LMT) This theory is first presented in Bresnan & Kanerva 1989 (see also Bresnan 2001a, chapter 14). In LMT, grammatical functions are treated as bundles of features, the two features in question being ±r(estricted) and ±o(bjective). These features generate a four-celled matrix, giving the four GFs recognised by LFG: -r 63. -o +o +r SUBJ OBLθ OBJ OBJθ It is an implicit claim of this classification that GFs fall into natural classes. Thus, an unrestricted GF need not have a semantic role; restricted GFs must have a semantic role. Expletive arguments are possible in subject and object position, but not elsewhere and these two GFs therefore constitute a natural class. A non-objective syntactic function is one that can be associated with an intransitive predicator. Subjects and obliques fulfil this condition and therefore also constitute a natural class. A further claim is that the four GFs can be ranked in a markedness hierarchy, with negative features taken as unmarked. SUBJ is negatively specified for both features and is therefore least marked; OBJθ is positively specified for both and is therefore most marked. The remaining GFs are ranked between these two: 64. Markedness of argument functions (Bresnan 2001a: 309, ex19) > OBJ, OBLθ > OBJθ SUBJ This hierarchy plays an important part in the mapping between argument structures and f-structures. Argument structure (a-structure) consists of a predicator and a list of the arguments associated with it in order of semantic prominence, where semantic prominence is equated with the thematic hierarchy generally accepted since Gruber (1965). The most prominent argument is designated the logical subject, and all arguments are annotated with a single feature value which is assigned on the basis of the semantics of the verb. Patientlike arguments receive the feature [-r] whilst other arguments receive the feature 21 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 [-o]. The feature assigned to secondary patients is subject to cross-linguistic variation, but in asymmetric object languages such as English and Indonesian it is [+o]. The logical subject is mapped to the subject function unless it is not the first item in the argument list, a situation which can occur in the case of verbs with athematic subjects. Other arguments are mapped to the lowest function on the hierarchy 64 which is compatible with the feature specification they carry. Comparing the operation of these principles for two English verbs with three arguments each will clarify this scheme. put has three arguments, of which only one, the second in thematic prominence, is a patient. Patientlike arguments are associated with the feature [-r], and this principle will apply to the second argument of put. Other arguments are associated with the feature [-o] by default, therefore the complete a-structure of the verb will be: 65. put < x, [-o] y, [-r] z > [-o] The logical subject is initial in this a-structure, and is therefore mapped onto SUBJ. The remaining roles are mapped to the most marked GFs compatible with their features: in the case of the second argument, this is OBJ, and in the case of the third argument it is OBLθ. The relevant part of the f-structure associated with the node under which the verb appears in a c-structure will have the following form: 66. PRED 'put < SUBJ, OBJ, OBLθ >' give differs in having two arguments which are patientlike. By the Asymmetrical Object Parameter (Bresnan & Moshi 1990), only one of these can be associated with the value [-r]. The secondary patient (the one that is least thematically prominent) is associated with the value [+o]. The resulting astructure is: 67. give < x, [-o] y, [-r] z> [+o] The first two arguments will be mapped to SUBJ and OBJ as before, but the feature associated with the third argument is compatible with the most marked GF, OBJθ and it is therefore mapped onto that GF. The resulting partial f-structure is: 68. PRED 'give < SUBJ, OBJ, OBJθ >' The licensing of arguments through the interaction of LMT, f-structure and cstructure will be discussed in section 1.4. 1.3.3 Functional categories in Indonesian syntax It was mentioned above that functional heads and their projections are part of the vocabulary of LFG c-structures, in common with most syntactic work since the 1980s. But such elements are rather weakly grammaticized in Indonesian. The language has no subject-verb or verb-object agreement, tense and aspect are usually not specified in clauses, number marking of nouns is 22 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 rare and definiteness is not marked obligatorily. These facts raise the question of whether it is appropriate to analyse the language as having functional categories, in particular IP as the category of clauses and DP as the category of referential expressions. The purpose of this section is to argue that such an analysis is plausible, and to show how functional categories are handled by LFG. The argument is easier to make in respect of the clause-level functional categories. Two types of element appear between the subject and the verb in Indonesian clauses, and neither of them can appear in other positions (with the same meaning). These two are negation: 69. Marisa, kamu tidak mengerti meN.understand Marisa 2SG NEG 'Marisa, you don't understand.' (SDM: 76) and expressions of modality: 70. Aku harus melihat buktinya dulu 1SG must meN.see proof.3 before 'I must see his evidence first.' (SDM: 76) Expressions of tense and/or aspect also appear in this position, but in at least some cases these words can be used in other clausal positions with similar meaning: 71. sedang membaca ketika saya Ia 3SG PROG meN.read while 1SG 'He was reading when I came.' (E&S: 487) 72. Sedang ia mengucapkan kata-katanya itu isterinya while 3SG meN.say word.DUP.3 that wife.3 menjerit meN.scream 'While he was uttering those words, his wife screamed.' (E&S: 487) datang come The case of sudah is similar: it is used to mark a completed action, but it can also be used as an adverb meaning 'already'16. Other adverbs can be placed between the subject and the verb: 73. tetap Proyek itu akan menjadi milik project that certainly FUT meN.become property Candra Surya Abadi sign Surya Abadi 'That project will certainly become the property of the Surya Abadi group.' (SDM: 76) This ambiguity is common in Western Austronesian languages. In Sasak, auxiliary verbs are distinguished positionally and by being clitic hosts, and the class includes wah which indicates past action. The same word appears in other positions, and then is translated as 'already' (Austin 2000). 16 23 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 Therefore the status of the temporal and aspectual markers is not entirely clearcut. However, akan, glossed FUT above, is clearly an auxiliary verb: it does not occur in other positions17 and it can be the base of a derived verb meng/di-akan 'aim, strive for' and a nominalisation keakanan 'the future'. On this basis, it is reasonable to suppose that there is a position above VP for a head expressing tense and/or aspect, and that other words like sudah and sedang have dual lexical entries both as adverbs and as functional items which can occupy that position. Various combinations of the three possibilities, including all three, are possible: 74. Partai itu tidak akan bisa membentuk able meN.form party that NEG FUT 'That party will not be able to form a government.' (IRG: 204) pemerintahan government This suggests that a series of functional head positions is required, as suggested first by Pollock (1989). This point will be important in the argumentation of section 2.2.1.2, but in general I will only include a single projection, labelled IP, in c-structure representations of clauses. There are two common ways of indicating definiteness in Indonesian nominals. The first is to use the demonstrative itu 'that' to indicate that the referent has been mentioned recently. The second is to use the third person possessive clitic -nya to indicate that the referent can be understood within the context of interaction, but has not previously been mentioned (IRG: 150151, Sukamto 1999). The contrast can be seen in the following examples: 75. Ibu sudah memasak nasi. Nasi itu di mother PERF meN.cook rice rice that LOC lemari cupboard 'Mother has cooked rice. It (LIT: that rice) is in the pantry.' 76. Kalau mau makan, nasinya di lemari cupboard if want eat rice.3 LOC 'If you want to eat, the rice is in the pantry.' (IRG: 151) The possessive clitic has the same distribution in this usage as it does in its use as a true possessive. This includes appearing closer to the head than demonstratives and relative clauses, and I therefore conclude that it cannot be of the category D. The case of the demonstrative is less clear. itu and its complementary ini 'this, these' are always the last element in a nominal phrase. This can lead to ambiguity in complex phrases: 77. 17 Anak yang naik sepeda itu tinggal dekat climb bike that live close child REL 'That child riding a bike lives near me.' OR 'The child riding that bike lives near me.' (IRG: 157) There is a homophonous preposition, which is not relevant here. 24 saya 1SG NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 I assume that appearing at the boundary of the phrase (at least most of the time) will be a characteristic of D, therefore these demonstratives are possible candidates. There is a strong reason for doubting that this is the correct analysis, however. It is a reliable generalisation that heads precede complements and other dependents in Indonesian. This is true for verbs and prepositions as can be seen in the preceding examples. It is also true for nouns: 78. dongeng tentang seorang legend about one.CLASS 'a story about a haji' (IRG: 150) haji haji and for adjectives: 79. Sulit untuk kita memperoleh difficult for 1PL.INCL meN.obtain 'It is difficult for us to get proof.' (IRG: 183) bukti proof And if the position argued for immediately above is correct, it is also true for the functional heads above VP. Therefore, it would be surprising if D were to take its complement in a right-branching structure18. On this basis, the only plausible candidates for the category D must occur before the head noun in nominal phrases. This condition restricts the candidates to quantifiers, including numbers, and classifiers. Classifiers can be eliminated from consideration, because they can only ever occur with a number (IRG: 134): 80. dua ekor two CLASS 'two horses' 81. seekor one.CLASS 'a horse' kuda horse kuda horse All Indonesian classifiers are originally nouns, and where there is no number associated with them, they revert to their nominal meaning: 82. ekor kuda tail horse 'horse's tail' This leaves, by elimination, the result that quantifiers are exponents of the category D in Indonesian. Although this hypothesis is semantically plausible, it is also not without syntactic problems. Chung (1976a) discusses the possibility of quantifiers shifting from the left edge of their nominal phrases to the right edge: Guilfoyle, Hung and Travis (1992: 401) state that: 'There is independent evidence that the determiner (sic) itu is only a modifier, not a functional head.' but without giving any details. I assume that what they have in mind is some version of the argument just given. 18 25 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 83. Semua pemain musik all player music 'All the musicians left early.' pulang go.home pagi morning 84. Pemain musik semua pulang player music all go.home 'All the musicians left early.' pagi morning She notes that this phenomenon is distinct from quantifier float, and that it is associated with some difference in meaning which she is unable to make specific. It should be noted that this possibility also exists for numbers and number/classifier combinations: 85. Dia mengimpor kursi sebanyak 8000 buah 8000 CLASS 3SG meN.import chair as.many 'He imported as many as 8000 chairs.' (IRG: 140) However, it is not possible in any of these cases to insert the quantificational element(s) within the remainder of the nominal phrase; the quantifier always must be on one edge of the structure. This is not counter-evidence to the hypothesis proposed, if we assume that the ordering of D and its complement is not rigid. In chapter 2, I assume the idea originally proposed by Postal (1969) that pronouns are determiners rather than nouns. In association with the analysis argued for above, this assumption predicts that pronouns and quantifiers should not co-occur. There are some possibilities for quantifiers to occur with pronouns, but they are limited in such a way that they are not counterevidence to the proposed analysis. The only numbers that can occur with pronouns are morphologically complex forms. Bare numbers cannot appear in this environment, only numbers prefixed with ber-: 86. dua orang two person 'two people' 87. *dua mereka two 3PL (FOR: 'two of them') 88. mereka berdua ber.two 3PL 'both of them' and these numbers do not occur with other nouns (IRG: 58). As discussed previously, morphology in Indonesian is derivational; ber- is a prefix commonly associated with intransitive verbs and adjectives. These facts, and its position in structures such as 88, suggest strongly that it is a modifier of some sort in this usage. Some quantifiers can occur with pronouns also. These words precede nouns, except in the case of quantifier shifting discussed above, but they must follow pronouns (IRG: 169-170): 26 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 89. Kami semua harus membuka jalan 1PL.INCL all must meN.open road 'All of us must clear a road first.' (IRG: 170) 90. *Semua kami..... dahulu before Whatever the correct analysis of quantifier shifting may be, it is significant that it is obligatory in just this case. The natural interpretation is that it is a mechanism of last resort here which rescues a structure predicted to be impossible on other grounds. Therefore, in the two situations in which pronouns can co-occur with quantifiers, the quantifier does not behave as it does with other nouns. It is plausible to maintain the assumption that quantifiers are generated as functional heads which take NP as their complement, but that the relative ordering of the head and its complement can be reversed under conditions which are not understood19. In what follows, I will treat the maximal projection of referential phrases as DP. I have argued that I and D are functional categories that exist in Indonesian. Following the standard account of such categories, this means that they are compulsory: every clause must be headed by an I and every referential phrase must be headed by D. But as mentioned at the start of this section, functional categories are commonly not instantiated in Indonesian. LFG does not allow null heads in c-structure, therefore it is appropriate to at least sketch here the approach taken to such a situation in the framework. A functional head and the lexical head of its complement map to the same functional structure. Where a functional head is obligatory (for example, if the grammar includes a rule which makes I(nflection) the head of a finite clause), and the functional head is realised morphologically, lexical integrity forces the lexical head to appear in the c-structure position of the functional head. This analysis accounts for many of the structures analysed as examples of head movement in Principles and Parameters accounts (see for example Pollock 1989, Grimshaw 1997). An example of the LFG analysis is the case of socalled verbal-nouns in Welsh. The finite verb is initial in Welsh, giving VSO order when a finite verb is used. But where a finite auxiliary is initial and a non-finite “verbal noun” appears, the word order is Aux-SVO: 91. Gwelodd Siôn ddraig saw. John dragon ‘John saw a dragon.’ 92. Gwnaeth Siôn weld draig John See.VN dragon do.3SG.PAST ‘John saw a dragon.’ (Sproat 1985: 176) We can assume that the c-structure of example 92 is that of an IP in which I takes a S consisting of an NP and a VP as complement (see Bresnan 2001a: 127-129 for argumentation). Example 91 can have the identical structure, with The question of whether any part of a structure such as sebanyak 8000 buah should be analysed as the specifier of D, and if so, what linear ordering principles apply is ignored here. Clearly, this an area which needs further research. 19 27 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 the sole exception that the Vo position in VP is not present. As the main verb in this example is finite, it must appear clause initially, that is as head of IP. By the structure-function correspondence principles previously discussed, the nodes associated with I, IP, V and VP all map to the same f-structure. Therefore, if lexical material were inserted under V in example 91, the PRED attribute of that f-structure would have two values, a possibility barred by general principles. Therefore only a V node dominating nothing is possible. But such node is not permitted by economy of expression, and it must be omitted all together. The resulting structures are the following (see Bresnan 2001a:129-131 for details): 93. 94. (Bresnan 2001a: 131, ex12) The c-structure head in I is then said to be an extended head or the co-head of the two c-structure categories (Bresnan 2001a: chapter 7). In the same way, where there is no phonological material realising an obligatory functional category, the lexical head will appear in the c-structure position of the functional head. I propose that, for Indonesian, such structures are associated with a radically underspecified value in the f-structure which is compatible with, and is given content by available contextual information. I will in general omit this attribute from f-structure representations, but it is shown in the following example with the value ( ... ) assigned to the attributes TAM (tense/aspect/mood) in the outer f-structure and DEF (definiteness) in the fstructure associated with the DP20,21: 20 Some non-branching nodes are omitted from the following c-structure. 28 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 95. The Indonesian facts perhaps appear closer to those of English, given the similarity of basic word order for the two languages. The accepted LFG analysis of finite clauses in English without auxiliaries is that they are subject-predicate structures, NP, XP, where XP is a VP (Bresnan 2001a: 115). A similar analysis does not seem possible for Indonesian, however, because IP is possible as the XP part of a subject-predicate structure with a non-verbal predicate: 21 I Minuman itu harus dingin Drink that must cold ‘The drinks must be cold.’ (IRG: 235) There is no obvious syntactic distinction between the subject of such clauses and subjects which I analyse as specifiers of IP. Therefore there would seem to be a large class of cases for which two analyses are possible: IP only, or NP-XP (= IP). 29 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 96. 97. Guru itu menulis buku tentang teacher that meN.write book about 'The teacher wrote a book about syntax.' sintaksis syntax I will not pursue this issue further, but I would suggest that the fact that the two functional categories both behave in this way is an important typological fact about Indonesian. Additional theoretical material relating to LFG is introduced in various parts of this work, as it is required. 1.4 The licensing of arguments We are now in a position to give a detailed statement of how LFG models the licensing of arguments in a configurational language, and to draw some implications from it. These implications will then be compared with data from Indonesian in order to reveal the nature of the problems to be studied in the following chapters. A verb has a list of semantic roles associated with it in its lexical entry. LMT provides a set of principles which associate syntactic features with these roles, and the result is an a-structure representation. The mapping principles of LMT then operate on this a-structure and the result is a partial f-structure. These steps are illustrated again for the English verb put: 98. put V < agent, theme, location > 99. put < x, [-o] y, [-r] 100. PRED 'put < SUBJ, OBJ, OBLθ z> [-o] (lexical entry) (a-structure) >' (partial f-structure) In order that the completeness condition be satisfied, the complete f-structure for a clause with this verb must contain a SUBJ, an OBJ and an OBLθ, and appropriate c-structure positions for them must exist. The X-bar rules of the language and the principles which provide functional annotations on c30 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 structure nodes must combine to fulfil this condition. I will assume an X-bar theory which is restricted to binary branching, except in the case of second objects, and the annotation principles given by Bresnan (2001a: 102-103). These principles state that the specifier of a functional category is a discourse function. In the case of IP, this allows the grammaticalized discourse function SUBJ to appear as its specifier. The principles also state that the complement of a lexical category is a non-discourse argument function, and this allows the GF OBJ to appear as complement of a verbal head. I assume that ditransitive verbs have two complement positions, their first V' is a ternary-branching node, and that two argument functions can appear in that case. Finally, a principle allows that a maximal projection adjoined to a constituent is optionally a nonargument function. These principles capture the generalisation that arguments are licensed structurally in two configurations: specifier of a functional category, and complement of a lexical category. The last principle also captures the generalisation that obliques and adjuncts have in common the property of not being licensed in this way. The observed properties of a language such as English fall out naturally from this account. SUBJ, OBJ and OBJθ are structurally licensed, OBLθ and adjuncts are not. As adjuncts are not assigned semantic roles by the verb, it is also predictable that the licensing mechanism will make a semantic contribution. Considerations of economy predict that the mechanism used to license non-terms will be identical for both types, and therefore it is predictable that the licensing mechanism for OBLθ will introduce semantic content also. If Indonesian is a configurational language without case-marking, the account of the licensing of arguments sketched above will apply, and we can make various predictions. We predict that there will be at least one nonsubject argument position where a DP is licensed structurally, and that where only one of these positions exists the DP which occupies it will have the GF OBJ. We predict that other non-subject arguments will appear in different positions and will require some other licensing mechanism, such as prepositions. And we predict that the licensing prepositions will code semantic distinctions between such arguments. The facts of the language cast doubt on all of these predictions. The first two predictions are thrown into question by data regarding the most common two argument clause types of the language: 101. Ali melihat gadis Ali meN.see girl 'Ali saw the girl.' itu that 102. Gadis girl itu that dilihat di.see Ali Ali 103. Gadis girl itu that dilihatnya di.see.3 104. Gadis itu dilihat kemarin oleh girl that di.see yesterday by 'The girl was seen (yesterday) by Ali / him.' 31 Ali Ali NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 Example 101 is a clause of the type described as active in most grammars (e.g. IRG), while examples 102-104 are described as passive. The subject precedes the verb in all cases. The phrase gadis itu is the object in example 101 according to the account developed above, but I have already noted that in a clause which differs only in lacking the verb prefix, this argument has different properties (see discussion of examples 52-56 in section 1.2.3). This is unexpected: we would predict that arguments which are exponents of the same GF will have similar properties. What of the postverbal elements in examples 102-104? In example 104, there is a prepositional phrase, which would be predicted by most theories of voice, and its function is marked by the preposition oleh 'by' just as predicted. But example 102 and example 103 do not fit into the picture so easily. We have a bare DP in one case and an attached pronoun in the other: neither has a preposition indicating its function in the clause. Rather than obliques, these have the coding properties of direct arguments, and therefore they are also objects by the theory given above. This in turn would mean that these are not passive clauses under commonly accepted assumptions about the nature of voice alternations. I will argue that these clauses present a more complicated picture still: there is good evidence to show that the post-verbal NP in example 102 is an oblique and that it is licensed by adjacency to the verb rather than by a preposition; there is also evidence that the pronoun in example 103 is a direct argument of the verb, and that it is not licensed by virtue of its syntactic position but by a morphological process. Example 102 already shows that prepositions are not necessary to code functional distinctions between non-terms, but another clause type shows that, in one environment, there is no relation between prepositions and functional distinctions at all. In this case, a variety of prepositions can be used to license a non-subject argument without any associated meaning distinction: 105. Saya tak ingat akan remember about 1SG NEG 'I don't remember you.' (E&S: 223) tuan sir 106. Masih ingatkah Anda pada 2SG to still remember.EMPH 'You still remember me?' (SDPGS: 108) 107. Saya jadi ingat dengan hit Scot Mackenzie 1SG thus remember with hit Scot Mackenzie 'Therefore I remember Scot Mackenzie's hit.' (McKay) saya? 1SG In addition, the preposition can be omitted altogether22: 22 For reasons discussed in section 4.1.2.3, this example is ambiguous as to clause type. 32 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 108. Di LOC mana where pun, PRT dalam inside keadaan situation pun, PRT dia 3SG selalu always ingat Marisa remember Marisa 'No matter where, no matter what situation, he always remembered Marisa.' (SDM: 265) All of these examples show that the licensing of non-subject arguments in Indonesian is not as straightforward as the theory given above would lead us to expect. The conclusion that will be argued for in this thesis is that there are two separate systems of clausal organization in Indonesian, each of which licenses non-subject arguments in a different way. Consequently, the properties of the arguments are different in each case. Clause types such as those seen in examples 101-104, and in some other types not yet introduced, will be argued to be typical of one system. I will suggest that this system is related to, though not identical with, the clausal syntax of other more conservative Austronesian languages. The characteristics of this system include that the predicate of the clause can be headed by all major categories, that terms within the predicate constituent are not available to syntactic processes such as extraction, and that there is a tendency for arguments internal to the predicate phrase to be joined to the head in a close relationship. The clause type seen in example 108, on the other hand, is typical of the other, more innovative system. The characteristics of this system include that the second major constituent of the clause must be headed by a verb and that the verb phrase is open to processes such as extraction. The licensing of arguments in the second system proposed above can be accounted for in a theory such as that outlined at the start of this section, but this is not the case for the first system. Additional assumptions are required in order to give an account of the licensing of non-subject arguments when the major constituents of the clause are subject and predicate, as opposed to subject and verb phrase, and I will argue that these additional assumptions are not compatible with LFG as currently formulated. 1.5 The origin of Bahasa Indonesia and the data used The Indonesian archipelago is an ethnically and linguistically diverse area. The vast majority of the languages spoken in the archipelago are from the Austronesian family, but the variety is nevertheless considerable. The feelings of distinct ethnic identity felt by groups in different regions are evidenced clearly in the recent political history of the Republic of Indonesia and the basis for the nation of Indonesia was a negative emotion. All the various groups involved in the nation-building movement were opposed to Dutch rule, but this did not necessarily imply any further solidarity between the groups. This was seen as a problem from early in the history of the independence movement, and the adoption of a national language was regarded as a crucial unifying move. The native language with the widest geographical distribution and use, although certainly not the greatest number 33 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 of speakers, was Malay, which had been the most common trade language of the region for centuries. This was adopted as the basis for a national language, Bahasa Indonesia ('language of Indonesia') by the nationalist youth movement in 1928, and became the official national language when the Republic of Indonesia was proclaimed in 1945. This language then became the official language of government and of education. The typical Indonesian of earlier generations learned her native language (for example Javanese or Balinese) at home and learned Indonesian at school. Indonesian was therefore a language without speakers for whom it was the first language, although many people could be said to have native speaker competence in the language, having commenced learning it at the age of five or six and having used it daily after that. More recently, increased mobility within the country has meant that marriages between people from different language backgrounds have become more common, and as a result there are now some speakers who have never learnt another language. An additional factor complicating matters is that, as previously mentioned, Malay was a trade language throughout the archipelago for a considerable period of time. Consequently, there are Malay dialects spoken at various places in Indonesia. For example, there are local dialects of Malay spoken in the widely separated cities of Jakarta (on Java) and Ambon (on Seram). There is also variation in the Indonesian spoken in different regions due to the influence of the different first languages spoken (Moeliono 1994, Tanner 1967). The result of this situation is that there are many varieties spoken which speakers will identify as Indonesian but which may vary widely from one to another. There is a prescriptive standard which is taught in schools and is used as the language of the government and the legal system, but is scarcely heard or read elsewhere. This is the variety which is described by most grammars (e.g. McDonald & Darjowidjojo 1967, IRG), but as a language in use Indonesian is richer. All the resources of the standard are available (if not actually used), but many other possibilities exist. To take a simple example, IRG has extensive discussion of the morphological causative formed with the suffix -kan (pp70-78), but does not mention the periphrastic causative formed with the verb kasi 'give'. This construction is recognised by Echols and Shadily's dictionary (1994): 109. Kasi lihat bukunya give see book.3 'Let me see the book.' (E&S: 263) although at the head of the entry for kasi they mark it as 'colloquial'23. Similar points are made with detailed evidence in respect of the so-called 'bare active' clause type by Voskuil (1996, appendix to chapter 8). Any study of Indonesian which restricts itself to the standard language will therefore miss important phenomena and valuable evidence. Kana (1986: 77-80) discusses this construction and suggests that it is characteristically eastern Indonesian. 23 34 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 The strategy adopted in the present study has been to use textual surveys to establish the scope of the phenomena under investigation in actual language use, and then to use the information obtained from text examples as the basis for focussed elicitation from native speakers. This is particularly true of the material presented in chapters 3 and 4: chapter 3 takes as starting point several published studies which are themselves based on text surveys and uses them as the basis for further investigation, while chapter 4 is based on my own survey of several texts (full details are given in section 4.1.2.1). Chapter 2 discusses aspects of grammar which are more central, and about which published information is generally reliable. Both the texts used and the judgments provided by my consultants reflect the careful usage of educated Indonesians. Most of the native speakers consulted are of Javanese origin and their judgments might not be replicated by speakers with another first language. The issue of L1 influence on Indonesian is certainly beyond the scope of this study (see Moeliono 1994, Steinhauer 1994). Bahasa Melayu, the national language of Malaysia, is also based on Malay and is essentially the same as Indonesian. The orthographies of the two languages were not identical, however, as they reflected the phonology to orthography mappings of different European languages. In 1972 the two nations adopted a uniform orthography, which is the current standard for both languages. Some examples in this work are taken from earlier sources, but the orthography has been standardised throughout, with the exception of proper names. 1.6 Overview of the remaining chapters CHAPTER 2 presents an analysis of the major two argument verb class of Indonesian. The data presented above has shown that there are four clause types in which such verbs can appear (examples 101, 102-104, 16, and 52). Two of these clause types have the more agentive argument as subject, and two have the more patientive argument as subject. In addition, there are restrictions on the nature of the non-subject argument in these last two clause types. These facts have never been explained in a straightforward fashion. The focus of the current study on the properties of non-subject arguments sheds some new light on the problems. CHAPTER 3 deals with a class of verbs usually described as adversative verbs. These are characterised formally by the presence of the circumfix ke- an around the root, and semantically (in almost all cases) by the subject of the verb experiencing adverse consequences as a result of the event: 110. Tomo kecurian uangnya Tomo ke.steal.an money.3 'Tomo has had his money stolen.' The non-subject argument in this clause type can be a bare DP as in 110, but sometimes it can be introduced by the preposition oleh 'by'. I will argue that this argument is of the same type as the agentive argument in some of the patient subject transitive clauses (such as example 102), and that in each case the argument is an oblique which is licensed by being adjoined to the verb. The ke- -an verbs are of interest also because they are naturally treated as 35 NON-SUBJECT ARGUMENTS IN INDONESIAN - CHAPTER 1 examples of predicate composition: that is, the circumfix should be analysed as a separate predicate which takes another predicate as one of its arguments. But unlike many other phenomena which can be analysed in this fashion, the way in which the governing predicate creates an argument structure is driven by semantic factors, not by the grammatical functions assigned by the governed predicate. The mapping from the a-structure of the complex predicate to grammatical functions also provides an argument in favour of the style of a-structure proposed in Manning (1996a). CHAPTER 4 is an analysis of the emotion and cognition verbs seen in examples 105-108. I argue that these words have multiple lexical entries, and that only the entries responsible for examples such as 108 are unequivocally verbal. The properties of the non-subject argument are such that it must be considered a direct argument, but other properties of the clause type also make it clear that this type of verb is outside of the transitive system analysed in chapter 2. This conflict forces a choice between two conclusions: either the non-subject argument of emotion verbs is a second object (an OBJθ in LFG), a conclusion which is theoretically and empirically problematic, or the assumption that the complex system studied in chapter 2 is the basic transitive verbal system of the language is wrong. On the basis of striking similarities between the emotion verb clauses and the unprefixed agent subject transitive clause type, the second conclusion is adopted. CHAPTER 5 attempts to motivate the conclusion reached in chapter 4 by an examination of the properties of the two different clausal systems which I propose exist in Indonesian. The interaction of extraction possibilities and the positions that can be occupied by enclitic pronouns is central in this analysis, with the generalisation concerning the distribution of enclitics most naturally stated in terms of the category predicate. On the basis of the properties of their non-subject arguments, as analysed in previous chapters, and on the basis of additional evidence, I propose that prefixed verbs and their projections fall within this category and unprefixed verbs do not. Comparison with other Austronesian languages and with proposed reconstructions of Proto-Austronesian syntax shows that the subject-predicate structure is conservative and that the other appears to be an innovation. The chapter concludes by arguing that this insight into the structure of Indonesian resists expression using the current resources of LFG, and suggests what sort of modifications might be necessary. 36 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 2 Indonesian Transitive Clauses Transitive verbs are those which take two term arguments, which in LFG will always be a SUBJ and an OBJ. The class of verbs traditionally treated as transitive in Indonesian can appear in four possible types of clause. In this chapter, I will assume that the possibility of appearing in all four types is the defining characteristic of transitive verbs in the language, and I investigate the status of the non-subject argument in each clause type. I will argue that three clause types always have two direct arguments, and the fourth sometimes does also. The patterns of linking between semantic roles and grammatical functions are not the same for all types, and the word orders associated with the different types vary, therefore this variety cannot obviously be treated within standard views of linking theory or of phrase structure. The analysis developed here is that the variety of the clause patterns, their individual attributes, and the partial complementary distribution between two of them can all be attributed to morphological properties of the language. In particular, I argue that Indonesian makes extensive use of 'small constructions' (Poser 1992), that is, transparent combinations of lexical items which must nevertheless be taken to be part of the morphological component of the language, and which are used as an extra-syntactic means of licensing term arguments. 2.1 The range of transitive constructions As mentioned above, I will assume that transitive verbs in Indonesian are defined by the possibility of appearing in four different types of clause. Many verbs which have two term arguments can take the prefix meN- (where N indicates a morphophonemically conditioned nasal consonant, dropped before liquids, see section 1.2.1for details): 1. Dia membaca buku 3SG meN.read book S/he read the book. itu that With a third person actor1, an alternative prefix di- can be used; word order shifts from actor-V-undergoer to undergoer-V-actor and the preposition oleh may precede the actor: 2. Buku itu dibaca (oleh) book that di.read (by) The book was read by Ali. Ali Ali In this type of construction, the third person actor often appears as the bound pronoun -nya encliticized to the verb: I use the terms actor and undergoer, from Role and Reference Grammar (Foley and Van Valin 1984, Van Valin & LaPolla 1997) as shorthand for respectively the more agent-like and the more patient-like arguments of a transitive clause. 1 37 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 3. Buku itu dibacanya book that di.read.3SG The book was read by him. For many speakers (although not all - see below), the construction exemplified in 2 is not possible with a first or second person actor. In such cases, another construction is used: 4. Buku itu saya book that 1SG The book, I read. baca read In this construction, the actor can only be a pronoun (or an address term used as a pronoun substitute), and first and second person singular pronouns are often procliticized to the verb: 5. Buku itu kubaca book that 1SG.read The book, I read. The clause types seen in examples 2-5 are all traditionally treated as passives. Finally, in spoken Indonesian it is common to hear sentences such as: 6. Dia baca buku book 3SG read S/he read the book. itu that Many grammars state that this construction is substandard (e.g. IRG: 'In colloquial usage meN- may frequently be dropped .... however ..... its occurrence is required in most natural contexts' [67-68]), or even ungrammatical (e.g. Winstedt 1913: 64), but it is used extensively2 and it has syntactic properties which differ from the standard construction seen in example 1. In the remainder of this chapter I will use the following purely descriptive terminology: the type of clause exemplified in 1 will be referred to as a meN-V clause or construction, the various subtypes seen in examples 2 and 3 will be referred to as di-V-PP, di-V-DP and di-V-nya clauses (also generally as di-V clauses), both types seen in 4 and 5 will be referred to as Pro-V clauses, and the constructions exemplified in example 6 will be referred to as bare verb constructions. In addition, I will refer to meN-V and bare verb clauses collectively as actor subject clauses, and di-V and Pro-V clauses will be referred to collectively as undergoer subject clauses. While in general the possibility of appearing in the full range of these constructions is criterial for assigning a verb to the class of transitives, there are some exceptions and marginal cases which require comment. There are a number of common verbs which have an initial m-, including makan 'eat', E. Anderson (1983) reports a survey in which bare verb forms were used in 21% of clauses addressed to strangers, and 67% of clauses addressed to family members. These figures should not be taken too literally - it is impossible to decide in some cases whether a clause has a bare verb or is of the Pro-V type - but they do indicate that the clause type is common in informal speech. 2 38 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 minum 'drink', minta 'ask' and mohon 'request', which normally appear in actor subject clauses without a prefix. All of these verbs can take the di- prefix freely, and indeed can occur with a meN- prefix also, although the meaning is usually slightly different in such cases. Thus makan simply means 'eat', but E&S give 'consume, use up' as the gloss for memakan. Most scholars regard these forms as including a fossilized form of the actor subject prefix (Verhaar 1984a). There are also a number of verbs which regularly take two arguments, but only appear in one type of clause. Almost all of these verbs are in the domain of cognition, typical examples being mau 'want' and ingin 'wish'. This class of verbs is discussed in detail in chapter 3 of the present work. Indonesian has a small number of verbs which take three term arguments as a lexical property of the verb root, such as beri 'give', kasi(h) 'give' and ajar 'teach', and these can occur in all four clause types. There are also many two argument verbs which become three argument verbs when suffixed with -i or -kan3. Typical examples of these derivations are: 7. Ayah mengirim uang father meN.send money 'Father sent money to me.' 8. Ayah mengirimi father meN.send.APPL 'Father sent me money.' 9. Pelayan mengambil segelas air waiter meN.fetch one.glass water 'The waiter fetched a glass of water.' 10. Pelayan mengambilkan tamu segelas guest one.glass waiter meN.fetch.APPL 'The waiter fetched the guest a glass of water.' saya 1SG kepada to saya 1SG uang money air water As far as I am aware, it is an exceptionless generalization that all verbs derived with the suffixes -i and -kan can occur in all four of the clause types discussed above. I will therefore treat ditransitive verbs as a subclass of transitive verbs (as defined here). Finally, there is a number of intransitive verbs which are ill-formed without the prefix meN-. In some cases, the prefix attaches to nouns and adjectives4 and can be regarded as derivational. Examples are tepi 'side' → menepi 'move to the side' and merah 'red' → memerah 'redden, become red'. There is also a group of intransitive verbal roots which require the prefix, such as menangis 'cry' and melompat 'jump'. In all the cases just discussed, the prefixed verb takes only a single argument and the undergoer subject clause types are therefore not possible. As the prefix appears in clauses in which the 3 There are also many two argument verbs which include these suffixes in their form. See chapter 3 for discussion of the question of whether adjective is a lexical category in Indonesian. Here, the term is used pre-theoretically. 4 39 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 actor is subject, the most natural hypothesis would be that intransitive verbs with the meN- prefix are unergatives, verbs with only an actor argument. However, the correlation is not perfect. Kana (1986: 55) states that 'most of the prefixed verbs are unergative, while some of the prefixed verbs are unaccusative.' A crucial test for Kana is the effect the suffix -kan has when added to an intransitive verb5. The suffixed verb has an additional argument which is an actor in some cases and an undergoer in others. Kana (1986: 5568) takes the first class to be unaccusatives and the second to be unergatives. Intransitive verbs with the meN- prefix occur in both classes: 11. Anak itu menangis child that meN.cry 'The child is crying.' 12. Anak itu menangiskan kucing yang hilang child that meN.cry.APPL cat lost REL 'The child cried over the lost cat.' (Kana 1986: 61, ex39a) 13. Penyakit malaria menular di Irian Jaya illness malaria meN.spread LOC Irian Jaya 'Malaria is spreading in Irian Jaya.' (Kana 1986: 65, ex61b) 14. Nyamuk menularkan penyakit malaria mosquito meN.spread.CAUS illness malaria 'Mosquitoes spread malaria.' (Kana 1986: 65, ex61a) I leave open the question of whether any or all of these intransitive possibilities can be or ought to be included in the account of the prefix meNdeveloped below. The availability of so many constructions for transitive verbs raises a number of issues. Firstly, the linking of arguments to grammatical functions is not straightforward. There are two clause types with the actor argument linked to the subject grammatical function, and two with the undergoer linked to subject. At first sight, this suggests redundancy which should be explained. Also, what is apparently the same verb form, the bare verb, can occur in two types of clause, those seen in examples 4 and 6, with opposite linking6. This also requires explanation. Secondly, the constituency pattern of the Pro-V construction is different to that of all the others. This clause type has both arguments of the transitive verb before the verb7 unlike the other three clause types. This variation must be accounted for within a constrained theory of phrase structure. Finally, the constraints on the nature of the argument linked 5 For further discussion, see section 4.3.1.2. Both of these examples can be exemplars of both of these clause types, although there is a clear difference in intonation when either the subject is postposed in example 4, or the nonsubject is topicalized in example 6 (Chung 1978a). 6 This discussion assumes neutral intonation. The subject of a Pro-V clause can follow the verb with distinctive intonation (Chung 1978a). For discussion of topicalization from meN-V and di-V clauses, see below. 7 40 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 to subject for the undergoer subject constructions must be explained. The fact that pronouns are treated differently from other noun phrases suggests an explanation in terms of the animacy hierarchy. There is some evidence for animacy-based splits in the treatment of actors in Western Austronesian languages (Musgrave 2000), but the split in Indonesian will be shown not to coincide with the split between the two types of undergoer subject clause, and I will argue that the possible types of actors in di-V and Pro-V constructions are conditioned by morphological properties of the language. Before exploring these questions, it is first necessary to make it clear that the prefix di- is part of the same paradigm as the prefix meN-, that is, a verbal prefix, and that it is not a proclitic pronoun, as has sometimes been argued (Cartier 1979, Guilfoyle, Hung and Travis 1992, Shibatani 1985). 2.2 The Status of di2.2.1 di- as pronoun Pronominal clitics are possible in the Pro-V construction for first and second person singular actors, see example 5 above. The first and second person proclitics are ku- and kau- respectively. These are transparent reductions of the free pronouns aku and engkau. The normal third person singular free pronoun is dia, therefore there is some plausibility to the suggestion that what appears to be a prefix, di-, should instead be considered to be a proclitic. This line of argument has the advantage of reducing the two undergoer subject constructions to one: all undergoer subject clauses are ProV clauses on this account. A recent presentation of this view is that of Guilfoyle, Hung and Travis (1992). I summarise their position below, and then turn to the arguments against the analysis of di- as a proclitic. Firstly, various problems arise in assigning a phrase-structure position to di- if it is a pronoun, and secondly, how semantic roles are associated with syntactic arguments is problematic. In addition, the historical evidence does not support the view that di- derives from dia. 2.2.1.1 THE ARGUMENTS OF GUILFOYLE, HUNG AND TRAVIS In an influential paper, Schachter (1976) reported that Tagalog (Austronesian, Philippines) does not always assign to a single DP all the properties normally associated with subjects. Philippine-type languages allow a variety of non-actor subjects, and in such constructions the syntactic subject does not have all the properties that might be expected. For example, the subject is the only DP available for relativization in such clauses, but it is not the only possible antecedent for reflexives. Indeed, the subject can itself be a reflexive anteceded by the actor. Two DPs in a single clause have subject properties, and as Schachter points out, it is not obvious which should be called subject. Guilfoyle, Hung and Travis (1992, afterwards GHT) is an attempt to capture the split in subject properties noted by Schachter in a configurational model (Principles & Parameters Theory). The crucial assumption in this account is that the agent (or more exactly, the more agentive argument of a transitive verb, what I am calling actor) is base-generated in Spec of VP and 41 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 that this position can remain occupied even if some other argument moves to Spec of IP, the canonical subject position, at s-structure. Two structural subject positions exist, and can be filled simultaneously and this explains the fact that subject properties can be shared between two DPs. The s-structure position of DPs depends, as is usual in such analyses, on the case-marking properties of the verb and its associated functional heads. The analysis is originally motivated for Malagasy and Tagalog, where all verbs are affixed, the affixes varying with the semantic role of the subject DP. GHT claim that each affix case marks a DP in a different d-structure position8, and that whichever DP is left without case-marking moves to Spec of IP in order to satisfy the case filter (Chomsky 1981). Thus the Malagasy infix -an-, which identifies the actor as the subject of the clause, is analysed by GHT as an assigner of case to the undergoer. This leaves the actor with no case-marking, and it must move to Spec of IP9: Semantic roles and d-structure positions are assumed to be related in a straightforward way, along the lines of Baker's (1988) Uniformity of Theta Assignment Hypothesis. 8 I do not deal with the GHT's account of case-marking in my main discussion, but it is worth noting some problems with this aspect of the analysis. Firstly, it seems to claim that predicates in the languages discussed have no intrinsic case-assigning properties, a claim which is contrary to the assumptions of most grammatical theories (but see Foley 1998 for a related view). The claim also encounters empirical problems: for example Tagalog recent perfective verbs have no subject-identifying morphology, but they nevertheless license arguments, and the surface coding of those arguments is the same as that given to nonsubject core arguments of normal verbs (Kroeger 1993: 50). Secondly, the analysis involves additional assumptions that GHT do not make clear: that the Extended Projection Principle holds and that therefore something has to move to Spec of IP in any clause; and that there is some 'last-resort' case-assignment process that deals with obliques after subject selection. This second assumption is necessary because, in both Tagalog and Malagasy, NPs which would otherwise appear as obliques can become subject. If these NPs had inherent case, assigned by the oblique determiner sa in Tagalog for example, then no motivation for the movement to Spec of IP would exist. But any such rescue mechanism seems an ad hoc manoeuvre, particularly in Indonesian where obliques are introduced by prepositions with semantic value. Thirdly, GHT equate the Austronesian clause types which do not have an actor as subject with the passive of languages such as English. But the morphology which characterises these clauses assigns case in the GHT account, whereas the morphology which characterises passive in current accounts (e.g. Jaeggli 1986) is analysed as absorbing a thetarole and therefore preventing case from being assigned. Finally, verbal morphology of the Philippine type has traditionally been viewed as directly reflecting the NP selected as subject (the title of Blake (1906) is "Expression of case by the verb in Tagalog"). The GHT analysis is counter-intuitive in that the verbal morphology is analysed as having a case relation with some NP, but never with the subject. 9 42 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 15. (GHT: ex5a, dotted arrow indicates case-marking) In languages such as Tagalog and Malagasy, the claim that different affixes case mark DPs in different d-structure positions forcing movement to Spec of IP for other DPs is reasonably straightforward as all verbs (see footnote 9 for an exception) have affixes. However, in the case of Indonesian (and Bahasa Melayu, they are treated together by GHT), the analysis is made more complex by the existence of unaffixed verbs. The analysis for meN- verbs is straightforward: the prefix enables the verb to case mark the undergoer argument and the actor argument must move from Spec of VP to Spec of IP to receive case there, just as in the tree above, example 15, modulo the ordering under IP. GHT claim that a unified analysis is possible of the undergoer subject constructions, treating them all as passives in which the actor argument is generated as Dº in the position Spec of VP, and is licensed (case-marked) by being adjoined to Vº, which has moved to I: 16. When the actor is first or second person, Dº is realized as either the full pronoun, or as a proclitic. When the actor is third person, Dº is realized as either one of the pronouns dia or mereka, or as the proclitic di- which can be co-indexed with a post-verbal NP, the complement of the actor pronoun determiner. This DP remains in its original position and the surface order is a result of the movement of the verb to Iº. In all cases, the undergoer argument does not receive case in its d-structure position as sister of Vº, and must move 43 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 to Spec of IP to pass the case filter. On GHT's account, the proclitic di- is a realization of the features for third person (see discussion of what this might mean below), and this accounts for the restrictions on the use of non-third person agents with di- verbs (cf. IRG:249-250 especially footnote 2). GHT note that liberal speakers allow di- verbs with a non-third person agent introduced by oleh in clauses similar to example 2 above. Their theory predicts that such speakers are no longer using di- as a spellout of third person features. The morpheme is not a pronoun in this case, and a postverbal actor cannot therefore be licensed by being co-indexed with the pronoun di- adjoined to the verb in I. The prediction then is that such speakers will not allow any bare DP agents with di- verbs. GHT provide no data on this point, and the prediction is hard to test without examining a substantial corpus of spoken Indonesian as speakers reject di-V clauses with post-verbal actors such as oleh saya and such constructions are avoided in writing. In the absence of such data, I would however doubt the accuracy of the prediction for the reason that a very common use of di-V clauses is with a third person enclitic actor (see example 3), and it would be surprising to discover that GHT's liberal speakers did not use such constructions. GHT's analyses of both the clause they term active (meN-V clauses) and those they term passive (di-V and Pro-V clauses) are thrown into doubt by the existence of bare verb constructions, which are not considered by GHT. Clauses such as example 6 above show that Indonesian verbs can license (case-mark) an undergoer in the absence of the prefix meN-. This fact goes against the GHT account, where this prefix case-marks the undergoer in situ. If the bare verb can case-mark its internal argument, why is the prefix needed at all? A similar line of argument can be advanced against the GHT analysis of Pro-V clauses: if a bare verb can case-mark the undergoer inside VP, why does it have to move to Spec of IP when the actor is a proclitic? 2.2.1.2 THE PHRASE-STRUCTURE POSITION OF DIThere are two possible positions as to the status of the reduced pronouns ku- and kau-: either they are true clitics, occupying their own phrase structure positions and phonologically dependent on the verb to their right; or they are attached to the verbs in the lexicon, in the fashion argued for French object clitics by Miller (1992). No issue arises for the type of Pro-V clause in which the actor appears as a full pronoun because in such cases pronoun substitutes are freely available and these do not constitute a closed class. Indonesian speakers command multiple exponents of first and second person pronouns, but these can also be freely substituted with kin terms, address terms and proper names in certain contexts10. In the following text example, the speaker, who is (or believes himself to be) the hearer's uncle, uses the kin term oom 'uncle' as a pronoun substitute: GHT (p398) state that the actor of a Pro-V clause 'is restricted to being a pronoun, clitic or proper name'. This is not quite correct: proper names can only be used in this type of clause as pronoun substitutes referring to the speaker or the hearer. A proper name with 3rd person reference is not possible. 10 44 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 17. Oom tidak mau uncle NEG want cintai sekaligus love.APPL as.well.as 'I do not want to hurt (SDM: 113) merusak wanita yang meN.damage woman REL Oom puja uncle worship the woman I love and worship.' Oom uncle Where a proper name is used as a pronoun substitute, this can consist of more than one phonological word, a two part name for example, or a title and a name: 18. Buku itu sudah Prof Jones baca, bukan? Prof Jones read book that PERF NEG 'Have you read that book?' (addressed to Prof. Jones) Such possibilities seem to rule out an analysis in which all pronouns in the Pro-V construction are treated as phonologically dependent on the following verb, with the apparent cliticization of the reduced forms only an orthographic convention. The forms which are orthographically combined with the verb fail most of the standard tests for lexical attachment. They do not undergo or trigger morphophonological processes11, there are no arbitrary gaps in their distribution and there are no suppletive forms. The sole criterion that might suggest a lexical origin is that co-ordination of following verbs is noticeably worse for the reduced forms than for the free forms: 19. ?Buku book itu that saya 1SG beli buy dan and 20. *Buku book itu that kubeli 1SG .buy 21. Buku itu kubeli dan and book that 1SG .buy 'I bought and read the book.' dan and baca read baca read kubaca 1SG.read However, it is not clear that co-ordination is a reliable test for the status of clitics; even Miller's French data show inconsistencies (see Sadler 1998 for discussion). Therefore, I will assume that it is preferable to treat all actors in the Pro-V clause type in a unified fashion as clitics having their own distinct phrase structure positions12. GHT assume that all actors in Indonesian originate as specifiers of VP. This position is not case-marked and therefore the actor always has to move to some other position to satisfy the case filter. Where the verb is prefixed This test is not crucial. No Indonesian prefix with a final vowel shows any morphophonemic variation. 11 Some of the arguments developed regarding the position which di- then is considered to occupy do not apply if this assumption is incorrect. However, the other arguments developed in following sections do still apply. 12 45 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 with meN-, which case-marks the undergoer, the actor moves to the specifier of IP, the canonical subject position. Where there is no meN-, some other move is required. di- is a pronoun, therefore all undergoer subject clauses have a pronoun actor. Following Postal (1969), GHT treat pronouns as determiners, and the actor pronoun is therefore originally the head of the DP which fills the specifier position in VP. This head moves to Iº and by joining the verb there it satisfies the case-marking requirement, giving the surface order Pronoun - Verb: 22. (GHT example 34) If there is a post-verbal actor, then it is the NP complement of the pronoun Dº which remains in its original position. The claim that the pronoun-verb combination finishes in the position of I is not straightforward however. GHT do not give any examples in which anything that might be an overt I appears, nor do they discuss this possibility, but functional elements can appear between the subject and the verb in Indonesian. When this occurs, the position to be occupied by the pronounverb combination is problematic. Three types of functional element can appear between the subject and the verb in Indonesian: negation, temporal/aspectual markers and modals. Leaving aside the question of what category any of these should be assigned to, it is a fact that in Pro-V clauses, all of these precede the pronoun actor: o 23. Buku itu tidak akan bisa book that NEG possible FUT 'I will not be able to read the book.' saya 1SG baca read All these elements might be accommodated by assuming that IP is not a single category, but rather can be analysed as a series of functional heads (Tense, Negation etc,) and their projections (Pollock 1989). But to account for an example such as 23, GHT are committed to a claim that there is some additional functional head above VP in Indonesian which is never overt, and which is required for their analysis to succeed. It is hard to think of plausible candidates for such a head. Some agreement head (Agr) might appear to be a possibility, but on closer inspection the idea fails. The necessary head must be the lowest head in the I complex, and this position is not compatible with AgrS. AgrO could occupy this position, but GHT explicitly characterise this 46 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 construction as a passive, which presumably means that the verb is intransitive and therefore AgrO cannot be present13. Also it cannot be the case, on GHT's arguments, that in a case such as example 23 the verb remains in its d-structure position. If this were so, the relation between the actor pronoun and the verb would still be the d-structure configuration assumed by GHT, a configuration in which they argue case-marking requirements cannot be met. The position of the preverbal pronouns, including di-, is therefore not adequately accounted for by GHT. The position of the pronoun actor is a problem also for a theory which does not permit movement such as LFG. The problematic cases for such a theory, however, come in clauses without any functional projections. As the tree in 24 shows, where there is some functional head above VP, two specifier positions are available which can be occupied by the subject nominal and the actor pronoun: 24. anjing itu sudah dipukul Ali dog that PERF di.hit Ali In the absence of such a head, it is by no means obvious where the actor pronoun should be placed. The structure can be taken to be a VP only (see trees in 25 below), in which case only a single specifier position is available for one of the arguments. Whichever nominal is analysed as occupying the available specifier position, the other is left without a home: GHT argue that passivization has no effect on theta assignment, but that it does prevent the verb from assigning accusative case. My argument assumes that this position rules out the possibility of AgrO occurring. 13 47 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 25. a anjing dog b itu that kupukul 1SG.hit Another possibility is that the tree in example 25b is correct as regards the position of the subject and the verb, and that the actor pronoun is a complement of the verb in a left-branching structure. However, Indonesian has heads robustly preceding complements for all major categories and this analysis is therefore implausible. LFG also allows lexical heads to appear in the position of functional heads, leaving the projection of the lexical head headless in the phrase-structure tree (although not in the functional structure) (Bresnan 2001a: Chapter 7 and section 1.3.3). Such an analysis would allow the verb to take the pronoun actor as a complement, but would produce the wrong linear order: 26. In the case where there is no functional head above VP in the clause, a theory such as LFG can offer no account of the phrase structure of Pro-V clauses, if we assume that the pronoun actor occupies an independent node and we exclude the possibility of a null head and the possibility of two specifiers in a single projection. It should be noted that the argument made in this last paragraph is true for Pro-V clauses in general, whether or not di- is a pronoun. If di- is claimed to be a pronoun, then this is an argument against that position. If di- is not a pronoun (as I claim), then this is a more general problem for the analysis of Pro-V clauses, one to which I return below. There I will argue that the insight of GHT, that the pronoun actor and the verb are in a close relation, is accurate, but that the nature of the relationship is morphological rather than syntactic. 48 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 2.2.1.3 ACTOR DOUBLING As discussed in the previous section, GHT's analysis assumes that an actor occurring immediately after the verb in a di-V clause is the complement of the D head di- which has undergone head movement. This claim raises various problems which are discussed in this section. The discussion concentrates on the case in which the postverbal actor is a pronoun, as these are the cases in which the problems are most evident. However, I take it that similar problems would arise in the case of DP postverbal actors, if the analysis were pursued in sufficient detail. GHT assume that pronouns are generated as Do, and that the complement of D is NP: 27. buku itu dibaca lelaki itu book that di.read man that 'The book was read by the man.' (GHT example 36) These assumptions lead to the prediction that a postverbal actor cannot be a pronoun. The pronoun is a D, and projects a DP, while the D head which has moved to become di- before the verb requires a DP complement. This prediction is not true: the pronouns -nya (see example 3) and mereka are both possible as postverbal actors, as are proper names, which GHT also treat as Ds14: 28. Buku itu dibaca mereka book that di.read 3PL 'The book was read by them.' 29. Buku itu dibaca Ali book that di.read Ali 'The book was read by Ali.' One solution to this problem might be to argue that D as a functional projection can take a range of categories as its complement (following the GHT acknowledge the possibility of proper names as post-verbal actors (n.30) and suggest that they can be optionally analysed as NPs, but they ignore the case of pronouns. 14 49 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 reasoning of Abney 1987), and that one possible complement is another DP. This would allow two D heads to be generated in one maximal projection, but would raise another problem, that of how two referential expressions can be generated in a single DP. If the reference of the two is identical, the structure would be redundant and if the reference of the two is not identical, the semantics must be incoherent. GHT's analysis is already open to this criticism in that they go against the spirit of Postal's work on which they rely. A central point of Postal's (1969) paper is that the noun projection associated with a pronoun determiner can never appear precisely because the determiner is referential. The same criticism applies even more strongly if two Ds are posited. Another possible solution might be to claim that the pronoun di- is the true argument and that any postverbal actor is an adjunct linked to the pronoun only by coreference, that is to adapt the analysis of nonconfigurational languages proposed by Jelinek (1984). This solution also can be rejected. Firstly, Jelinek's proposal is intended to explain a cluster of properties which co-occur in the languages referred to as non-configurational, three of which are crucial: free word order, discontinuous constituents and null anaphora. Indonesian does not have free word order, nor does it allow discontinuous constituents. It is more tolerant of omitted arguments than English, but Jelinek (1984) argues with respect to Chinese that this phenomenon can occur in configurational languages and should be accounted for outside of sentence grammar. There is therefore no reason to assume that Indonesian is a systematically non-configurational language. Secondly, there is no reason to assume that it has non-configurational characteristics in just the one clause type. Two predictions which would be made by the proposed analysis fail, and one of these also undermines a strength of the GHT analysis as it stands. If di- is a true pronominal argument which can be optionally doubled by an adjunct, then it would be predicted that a clause with no other exponent of actor could have a reading with a specific third person actor. This is not true; such clauses have a reading similar to that of agentless passive clauses in English15: 30. Barang ini akan dikirim ke Jepang goods this FUT di.send to Japan 'These goods will be sent to Japan.' NOT: 'These goods will be sent to Japan by him/her/them.' Secondly, Jelinek's analysis makes it a characteristic of adjuncts doubling pronominal arguments that they have no fixed position in the clause (word order is free). If the postverbal actor were such an adjunct, it would be predicted that it could appear in any position. This is not true. The postverbal actor must be immediately adjacent to the verb, unless it is introduced by the preposition oleh. This fact is accounted for in GHT's analysis, but modifying In sequences of action clauses with the same actor, it is common in Indonesian discourse for the non-initial clauses to be di-V clauses without an actor. In such cases, the actor is specific but I assume that such possibilities are accounted for outside of sentence grammar as previously discussed. 15 50 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 their account in the fashion discussed here would lose this advantage as the postverbal actor would no longer be generated in Spec of VP, an A-position. I discuss the possibility of a similar approach in the LFG framework below, and reject it for similar reasons. The problem of actor doubling is equally awkward in other frameworks. For example, an analysis in LFG would be unable to assign a well-formed functional structure to a clause with a post-verbal actor. Whatever grammatical function is assigned to the pronoun di- should also be assigned to the postverbal actor for the interpretation to come out right. But LFG has a strict condition which bars a single attribute, such as a grammatical function, having two values. No coherent f-structure is possible as demonstrated for parallel facts from Irish in a detailed analysis by Andrews (1990). One possible solution in LFG would be to claim that the pronoun carries the grammatical function, and that the post-verbal actor is licensed by being the exponent of some discourse function (Bresnan and Mchombo 1987). However, the behaviour of the post-verbal actor is not at all what might be expected on this analysis: occurring in non-typical linear position, especially at the edge of a clause, is characteristic of discourse functional DPs, but a postverbal actor must always be adjacent to the verb. It can be clause-final, but this is accidental: it happens only when there is not other material in the clause to follow. It would also be unexpected for an argument assigned a discourse function to be expressed as a reduced pronoun. The reduced pronouns ku- and kau- behave exactly as might be expected given these considerations, no other exponent of the actor role can occur in the clause. But if di- is really a reduced form of the pronoun, parallel to ku- and kau-, then we would predict that it should also be possible for these pronouns to be doubled after the verb. All such possibilities are ungrammatical: 31. *Buku book itu that kubacaku 1SG.read.1SG 32. *Buku book itu that kubaca 1SG.read 33. *Buku itu kubaca book that 1SG.read (FOR: 'I read the book.') saya 1SG oleh by saya 1SG This might be accounted for on the grounds that the 1st and 2nd person pronouns are fully referential in context, whereas a 3rd person pronoun may need additional specification. However, this argument cannot be maintained because for all speakers -nya is a possible post-verbal actor but it is not more specific than the pronoun dia, the supposed source of di-. Canonically, dia has only human (or at least animate reference); some speakers allow it to have inanimate reference also, but in all such cases its reference is singular. For -nya on the other hand, most speakers allow inanimate reference and many allow plural reference. Therefore, if di- is a reduced form of dia, post-verbal -nya does not make the reference more specific. The fact that ku- and kau- do not allow a post-verbal actor must be attributed to some other reason. This 51 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 asymmetry between ku- and kau- and di- cannot be explained if di- is taken to be a true pronoun. It can be explained if di- has less referential content than the true pronouns, and I discuss this possibility below. Although examples 4 and 5 give two structures for the Pro-V clause type, and although the reduced pronoun is conventionally written as one word with the verb (these forms were often referred to formerly as 'conjugated verb forms', see for example Gonda 1949), there is little evidence of any syntactic difference between the two possibilities (apart from the co-ordination data discussed above). Therefore, we would also predict that dia as actor in a Pro-V clause could be doubled with a post-verbal actor. As pointed out by Arka and Manning (to appear), this prediction is also not correct (I omit the possibility of doubling with -nya which is ruled out below on other grounds): 34. *Orang man itu that dia 3SG lihat see 35. *Orang itu dia lihat man that 3SG see (FOR: 'Ali saw that person.') Ali Ali oleh by Ali Ali Thus in addition to the theoretical arguments for not analyzing di- as part of the same paradigm as ku- and kau-, there is direct morphosyntactic evidence to support this position. There is also evidence that di- has significantly similar properties to the verb prefix meN-, and this is discussed in the following section. One response to the evidence discussed in this section might be to claim that di- does indeed originate from dia and was at some stage a pronoun in its own right, but that it is now an agreement marker, that is, its referential content has become reduced. The split in actor possibilities between the two undergoer subject clause types would still be accounted for, and the arguments made above would be irrelevant. However, this possibility can also be eliminated. As mentioned previously, 1st and 2nd person actors do occur in di-V clauses on occasion, typically in prepositional phrases headed by oleh16. This already casts doubt on the interpretation of di- as an agreement marker, but further evidence renders it untenable. If di- were an agreement marker, we would predict that its effects would be most obvious in di-V clauses without an overt actor. But it is precisely in such cases that it is possible to have an implied 1st or 2nd person actor (Kaswanti Purwo 1988)17: 36. 16 Kau ini hanya cewek yang diurus 2SG this only girl di.take.care.of REL 'As for you here, it's only girls that you have on your mind' (LIT: 'You here, it's only girls that are taken care of.') (Hilman & Boim 1992: 80) Another possibility is discussed in section 2.4.3.2. Jelinek (1984) discusses examples of this type in non-configurational and pro-drop languages. Her discussion is not relevant to Indonesian, for the reasons previously given. 17 52 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 Therefore, the weaker hypothesis that di- is synchronically an agreement marker also fails. 2.2.1.4 SEMANTIC ROLE ASSIGNMENT In general, modern syntax assumes a bi-unique mapping between semantic roles and syntactic arguments. Where arguments have been advanced that this relationship should be loosened, this has typically been in the direction of allowing an argument to have more than one semantic role (e.g. Jackendoff 1987 and elsewhere). An analysis which claims that di- is a pronoun also claims that the opposite type of relationship is true: that more than one argument can have the same semantic role. Intuitively, this possibility seems more pernicious, and major syntactic theories have made greater efforts to exclude it. GHT do not discuss this issue, so I am forced to speculate how they might deal with it. The theta criterion must be satisfied at d-structure in the framework GHT adopt. This seems straightforward; the DP which is in the specifier of VP at d-structure is assigned the theta role Agent. But what happens at s-structure? It is perfectly possible for no post-verbal actor to appear in a di-V clause: 37. Barang ini akan dikirim ke goods this FUT di.send to 'These goods will be sent to Japan.' Jepang Japan This would suggest that the pronoun di- carries the theta role, and for a consistent analysis we would expect this to be the case even when there is a post-verbal actor. The fact that the pronoun is the head of the category to which the theta role was originally assigned might be taken as an additional argument for this position. But the post-verbal actor certainly contributes to the interpretation of the clause: what then is the relationship between it and the preverbal pronoun? Certainly the classic Principles and Parameters answer to this sort of problem is not available to GHT. The two positions which apparently are assigned the same theta role do not form a chain, the categories are not identical and the binding relations are not admissible. The argument applies a fortiori if a chain including the trace of D is considered. The question posed above is a reflex of the more general question raised above: how is it possible for two referential expressions to be generated under one DP node? The problem is equally awkward in other frameworks. For example, an analysis in LFG would be unable to assign a well-formed functional structure to a clause with a post-verbal actor. Whatever grammatical function is assigned to the pronoun di- should also be assigned to the post-verbal actor for the interpretation to come out right. But LFG has a strict condition which bars a single attribute, such as a grammatical function, having two values. No coherent f-structure is possible if di- is a pronoun, that is if it has a PRED value. This feature in LFG is part of the lexical entry of any item that has a semantic representation associated with it. For example, the clause in 53 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 example 3, Buku itu dibacanya, would be associated with an f-structure along the lines of 38, if we assume that di- is a pronoun18: 38. The object grammatical function has two values here, which is not permitted (Kaplan and Bresnan 1982: 181). The alternative is that one of the pronouns is not the value of any attribute in the f-structure: 39. This f-structure is equally impossible: if the pronoun is not assigned a function in some f-structure, it cannot be semantically interpreted. 2.2.1.5 HISTORICAL EVIDENCE The question of the historical origin of di- and the question of its synchronic analysis are logically independent. The prefix could have originated as a reduced form of the pronoun dia and have undergone reanalysis to a verb prefix at some point; equally it could have some other origin and be currently analysed by speakers as a pronoun (although the evidence presented here shows this to be highly improbable). However, as the proponents of the position that di- is a pronoun typically use the similarity in shape of the prefix and the free pronoun as an argument without adducing historical evidence (Kana 1986 is an honourable exception), it is appropriate to briefly consider the available material. This example uses -nya as post-verbal agent because it is unequivocally a term argument see discussion below, section 2.3.1.3. I later argue that in such cases, -nya is morphologically attached to the verb. The impossibility of the analysis is only more apparent in this case: the verb form includes a pronoun in its morphological representation and the function cannot take any other value (cf. a closely parallel analysis of Irish subject morphology, Andrews 1990). 18 54 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 The historical literature offers two sources besides dia for di-: the prefix ni- which was a passive or object-oriented marker in Old Malay, and the preposition di used as an agent marker. The first position has been argued by earlier linguists as reflected in De Casparis (1956) and by Hopper (1979). However, the Old Malay prefix ni- was compatible with first and second person actors as well as third persons. Therefore, if this is the source of di-, some account of how its usage came to be restricted must be provided, and this has not yet been done as far as I am aware. In addition, this proposal has to account for the shift from a nasal to an oral consonant. This can be done in the case of the Old Malay prefix mar-, which has as its contemporary reflex ber-, but the arguments which apply in that case do not apply to ni- (Adelaar 1992b, p.c.). The second source is better supported in the literature. In at least one Malayic dialect, Salako, it is still obvious that di is a pre-nominal marker of agency which can also precede the verb, and a similar phenomenon is attested in Minangkabau (Adelaar 1992a, b). Another argument against deriving di- from the free pronoun dia is offered by Wolff (cited by Kana 1986:102) who suggests that it entered Indonesian from Javanese19.The possible reasoning behind this suggestion leads to another issue. There are actually two 3rd person singular pronouns in Indonesian, dia and ia. The second of these only occurs in writing and is restricted to preverbal positions (simplifying slightly: see discussion in section 1.2.2), typically subject of a clause or agent in a Pro-V clause. The existence of both these forms is the remnant of an older system in which the personal pronouns had special oblique forms: daku for 1st person, dikau for 2nd person with dia as the 3rd person oblique form. So in older forms of the language, the ia form would have been used as actor of Pro-V clauses, making the derivation of di- from a free pronoun less plausible. Javanese has a set of verb prefixes used in undergoer subject clauses, and the prefix indicating 1st person actor is tak- or dak-. The origin of Wolff's suggestion therefore would seem to be an assumption that the initial consonant was more widespread in Javanese than in Indonesian, and that it is therefore a more plausible source for the Indonesian form20. In his monograph on Proto-Malayic, Adelaar (1992a:162) does not use di- as the basis for any reconstruction, although in later work (Adelaar 1992b) he accepts a prepositional origin for the prefix and it is possible to say with reasonable confidence that a pronominal origin for di- is unlikely. To the extent that this conclusion is relevant to synchronic analysis then, historical evidence agrees with the arguments presented here. 2.2.1.6 SUMMARY The preceding sections have shown that serious theoretical problems arise when di- is analysed as a pronoun, that there is direct morphosyntactic evidence that it behaves differently to the two reduced pronouns ku- and kau-, and that the historical evidence does not support the free pronoun dia as the 19 This discussion is necessarily speculative: Kana gives only the bare claim made by Wolff. 20 This point was made to me by Sander Adelaar (p.c.) 55 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 source of di-. There is, however, evidence that di- shares properties with the verb prefix meN-, and to I now turn to this. 2.2.2 di- as verb prefix This section will examine the morphosyntactic evidence that di- is a verb prefix parallel to meN-. Much of the evidence hinges on properties of postverbal arguments in clauses with prefixed verbs, and this introduces an additional complication. Recent work by Arka and Manning (to appear, afterwards A&M) has argued that the post-verbal actor in a di-V clause is only a term (a direct argument) when it takes the form of the attached pronoun -nya. Noun phrase post-verbal actors are non-terms. The evidence for this claim will be discussed in detail in section 2.3.1.3; the crucial fact is that -nya can antecede an anaphor in subject position of a di-V clause. But no other post-verbal actor can do this: 40. 41. *Dirinya self.3 tidak NEG diperhatikan di.look.after Haris Haris Dirinya tidak diperhatikannya self.3 di.look.after.3 NEG '(S)he didn't take care of her-/himself.' Accepting this data, and Arka and Manning's arguments, means also accepting that a direct comparison of the non-subject arguments of transitive verbs in meN-V and di-V clauses is not always legitimate. Therefore, where this is a consideration, I use the second object of a di-prefixed ditransitive verb as the comparison. Such arguments are terms on the basis of a separate test (quantifier float). 2.2.2.1 POSTVERBAL ATTACHED PRONOUNS Attached pronouns occur only with verbs with a prefix, meN- or di-. This generalisation is independent of the status of the non-subject argument because, as mentioned above, I assume that the attached pronoun is still a direct argument following a di- prefixed verb. With a meN- verb, the full range of attached pronouns is possible, -ku for 1st person singular, -mu for 2nd person and -nya for 3rd person: 42. Ali melihatku/-mu/-nya Ali meN.see.1SG/2/3 'Ali saw me/you/(s)he).' 43. dia tidak dapat melakukannya 3SG NEG able meN.act.APPL.3SG 'he couldn't do it again' (SDM:70) lagi further As seen above (example 3) and in the following text example, the pronoun -nya can attach to a di- verb also: 56 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 44. Tanah itu diwarisinya dari kakeknya land that di.heir.APPL.3SG from grandfather.3SG 'That land he inherited from his grandfather.' (SDM:245) The other two attached pronouns do not occur, as di-V constructions are restricted to 3rd person actors. In contrast, no bare verb form in Indonesian can have a pronoun attached to the right. In addition to the Pro-V clause type, there is also an actor subject bare verb clause type in Indonesian (introduced in example 6): 45. Ali baca buku Ali read book 'Ali read the book.' itu that This type of structure is common in spoken Indonesian and in writing which emulates colloquial speech, although it is not part of the prescriptive standard (see Voskuil 1996, Appendix to chapter 8 for discussion). The two types of clause with bare verb forms are distinguished by the positioning of temporal and modal auxiliaries which occur to the right of the actor in bare verb clauses and to its left in Pro-V clauses. An attached pronoun following the verb is impossible in both cases: 46. *Akan FUT 47. saya 1SG lihatnya see.3 *Saya akan lihatnya see.3 1SG FUT (FOR: 'I will see him.') Further evidence for this generalisation, that only prefixed verbs can host attached pronouns, comes from a class of quasi-verbal predicates. These are predicates which denote emotional and cognitive states and which do not take the prefixes meN- or di-. They normally take a single direct argument, the experiencer, with the stimulus appearing as a prepositional phrase. Some of these words also allow the possibility of the stimulus appearing as a bare DP, but it cannot be an attached pronoun unless a transitive verb is derived using an applicative suffix21: Pronouns can be attached to some prepositions, and therefore the stimulus can be an attached pronoun when it is within a prepositional phrase: Saya sayang padanya. Chapter 4 is an extensive discussion of this class of words. 21 57 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 48. Saya 1SG sayang pity pada to 49. Saya 1SG sayang pity Siti Siti 50. *Saya 1SG 51. Saya menyayanginya 1SG meN.pity.APPL.3 'I pity Siti / her.' Siti Siti sayangnya pity.3 Therefore, di- patterns with meN- in this respect: they both license an attached pronoun. I argue below (section 2.5.2) that the relation between the host verb and the pronoun varies depending on the prefix. This does not affect the descriptive generalisation discussed here, but it does suggest that it is a matter of phonological realization, rather than a syntactic property. 2.2.2.2 EXTRACTION AND TOPICALIZATION Both meN- and di- block extraction of a non-subject argument. This fact, combined with the fact that verbs with these prefixes are the most common in written Indonesian, has led to the received wisdom that Indonesian permits only subject extraction. This is incorrect; at least two types of clause allow extraction of the non-subject argument and crucially both of these have unprefixed verbs. The data in this section uses relative clauses throughout to exemplify the extraction possibilities. In the prescriptive standard, questions are formed as equational clauses with the question word as predicate and a headless relative clause as subject: 52. Siapa yang kauundang? who 2.invite REL 'Who did you invite?' (LIT: Who is the one you invited?') Therefore, the discussion of questions falls under the discussion of relative clauses for this register. There are also some wh- in situ possibilities which would needlessly complicate the discussion, and I ignore these here. A non-subject argument can head a relative clause with predicate construction: 53. orang yang saya 1SG person REL 'the person that I like' suka like an emotion itu that Similarly, a non-subject argument can be extracted with the bare verb actor subject construction previously discussed. But extraction of the actor from a Pro-V clause is totally impossible, giving rise to the following paradigm22: Example 47 is possible only if the name Badu is used as a pronoun substitute. Badu cannot be a 3rd person in this case; see discussion in section 2.2.1.2. 22 58 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 54. (Bare verb, subject extraction) Inilah Badu yang sudah baca this.PRT Badu REL read PERF 'This is Badu who has read the book.' 55. (Bare verb, non-subject extraction) Inilah buku yang Badu sudah this.PRT book REL Badu PERF 56. (Pro-V clause, subject extraction) Inilah buku yang sudah Badu this.PRT book REL Badu PERF 'This is the book that Badu has read.' 57. buku book itu that baca read baca read (Pro-V clause, non-subject extraction) *Inilah Badu yang buku itu sudah book that PERF this.PRT Badu REL (FOR: 'This is Badu who has read the book.') baca read These examples demonstrate that Indonesian, unlike say Tagalog, does not have a blanket ban on the extraction of non-subjects. Possible reasons for the impossibility of example 57 will be discussed below. Verbs prefixed with meN- cannot have their non-subject argument extracted: 58. *Itulah buku yang saya membaca that.PRT book REL 1SG meN.read (FOR: 'That is the book that I read.') The same apparently applies to di-V clauses: 59. *Ali yang buku itu book that Ali REL (FOR: 'Ali who read the book') dibaca di.read But I have argued above that bare DP actors in di-V clauses are not direct arguments of the verb, and there is evidence that this type of extraction is limited to direct arguments in Indonesian23. Therefore, the comparison between example 58 and example 59 is not legitimate. The crucial data relate to the possibility of extracting the second object of a ditransitive verb. Such objects can be shown to be direct arguments of the verb, and I will assume that they are VP constituents in both meN-V and di-V clauses. If verb prefixation makes VP an island for extraction, we would predict that such second objects would not be extractable for either type of clause, and this is the case: 60. 23 *buku yang Ali memberi saya book Ali meN.give 1SG REL (FOR: 'the book that Ali gave to me') See chapter 5 for discussion. 59 itu that NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 61. *buku yang saya diberi Ali itu book 1SG di.give Ali that REL (FOR: 'the book that was given to me by Ali') Again, the evidence suggests that verbs in di-V clauses behave in the same way as verbs prefixed with meN-. In addition to extraction being impossible from the environment of a meN- prefixed verb, topicalization also fails from such structures: 62. *Buku ini Badu sudah book this Badu PERF (FOR: 'This book, Badu read.') membaca meN.read As might be expected, given the requirement that bare DP actors appear adjacent to the verb, the identical topicalization from a di-V clause is also bad. But if oleh is used, the structure is perfectly acceptable: 63. *Badu Badu buku book ini this dibaca di.read 64. Oleh Badu buku ini dibaca by Badu book this di.read 'By Badu, this book was read.' The considerations previously discussed suggest that in this case also the true comparison should be with the second object of a ditransitive. Topicalization of such an argument is not possible from a clause with either a meN- verb or a di- verb: 65. *Buku itu Ali memberi book that Ali meN.give.3 (FOR: 'That book, Ali gave Siti.') Siti Siti 66. *Buku itu Siti diberi Ali book that Siti di.give Ali (FOR: 'That book, Siti was given by Ali.') Topicalization of the non-subject argument from a clause with a bare verb form is fully acceptable: 67. Film itu saya mau film that 1SG want 'That film I want to see.' 68. Gadis itu saya girl that 1SG 'That girl I like a lot.' suka like lihat see sekali very Topicalization is not possible for the non-subject argument of a Pro-V clause, but again I would suggest that this is for independent reasons to be discussed below. The contrast between bare verbs and verbs with prefixed morphemes is again the crucial point: meN- and di- behave in the same way. 60 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 Voskuil (1996: 192-195) claims that meN- and di- have an additional common syntactic property: they both license a zero anaphor as their nonsubject argument. He offers the evidence reviewed above of contrasts between the acceptability of left-dislocations from meN-V clauses and from parallel bare verb clauses to support his claim. However, when a headless relative clause is part of an equational predication, the constraint does not apply: 69. Surat ini saya yang menulis(nya) letter this 1SG meN.write REL 'As for this letter, it is me who wrote (it).' (Voskuil 1996: 193 ex.32) Voskuil claims that this construction is possible because it does not involve movement leaving a trace, but rather there is a null resumptive pronoun in the original position of the left-dislocated DP. Given this assumption, tests which distinguish between traces and null pronouns (pro) are available. Firstly, nonreferential DPs can be moved but cannot be left-dislocated, because pro is referential: 70. Tidak satu buku pun saya akan baca one book EMPH 1SG read FUT 'Not one single book I shall read.' (Voskuil 1996: 194 ex.37) NEG 71. *Tidak satu buku pun one book EMPH (Voskuil 1996: 194 ex.38) NEG saya 1SG yang REL menulis meN.write Secondly, locally bound anaphors can be moved but not dislocated because anaphors and DP traces are both subject to Principle A of the binding theory (Chomsky 1981), but pro is a pronoun subject to Principle B and therefore in complementary distribution with anaphors: 72. Dirinya sendiri Ali tidak percayai believe.APPL self.3 self Ali NEG 'Himself Ali doesn't trust.' (Voskuil 1996: 195 ex.39) 73. *Dirinya sendiri Ali yang self.3 self Ali REL (Voskuil 1996: 195 ex.40) mempercayai meN.believe.APPL All the evidence which Voskuil gives for these claims involves meN-V clauses, but the arguments developed here would lead us to predict that di-V clauses should behave in a parallel fashion. However, this is not the case. Such constructions with a di-prefixed verb in the relative clause are consistently bad: 74. *Ali surat ini yang ditulis di.write Ali letter this REL (FOR: 'As for Ali, it was this letter that was written (by him).') 61 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 75. *Ali surat ini yang ditulisnya Ali letter this REL di.write.3 (FOR: 'As for Ali, it was this letter that was written (by him).') Voskuil's claim that this possibility is in some sense dependent on the possibility of a resumptive pronoun means that the status of the actor should not be relevant here. The resumptive pronoun will always be a term, just as the undergoer of a meN-V clause is. Leaving aside this point, we note that in fact this construction is not possible also with the second object of a ditransitive verb: 76. *Buku itu Ali yang diberinya di.give.3 book that Ali REL (FOR: 'As for that book, it was Ali who was given it by her.') I will suggest a reason for this difference between meN- verbs and di- verbs below. 2.2.3 Summary The preceding sections have presented a variety of arguments designed to show that di- is not a proclitic pronoun, a reduced form of dia. The most detailed analysis along these lines, that of GHT, has been shown to be flawed. Both historical and morphosyntactic evidence have been presented to show that any such analysis is incorrect, and finally it has been shown that such an analysis is not compatible with basic principles of LFG. On the other hand, there is morphosyntactic evidence which shows that di- has properties in common with meN-, which is unequivocally a verb prefix24. I will therefore proceed on the basis that di- is a verbal prefix, in paradigmatic relation to meN- which has never been claimed to be anything but a prefix, to my knowledge. The next topic for discussion is then the syntactic function of these prefixes. 2.3 Prefixed Verbs: meN-V and di-V clauses 2.3.1 The status of non-subject actors The constructions with a di- prefixed verb are described as passives in many grammars, although reservations are often expressed. The following from IRG is typical: Some grammars of Indonesian do not use the terms active and passive, instead using terms such as subject and object construction or subjective Many Austronesian languages have a prefix indicating that the actor is the subject of a clause which consists of a nasal consonant only (e.g. Balinese, Sasak, Javanese). In informal usage, such a prefix is sometimes used in Indonesian. For example, the actor subject form of the verb tulis 'write' would be nulis rather than menulis. Umar Muslim (p.c.) points out that in this case, the verb behaves like the bare verbs discussed in the preceding sections, and not like a prefixed verb. This suggests that meN- is not a unitary morpheme at some level of analysis (see Benjamins 1997). 24 62 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 and objective focus25. This is because there are differences between the constructions in Indonesian and the active and passive in European languages such as English, both in structure and function. Nevertheless, there are also important similarities, and the relationship between the two constructions is often similar to the relationship between active and passive in English, allowing the same terms to be used to describe them. (p247) The differences commonly noted between this construction and the English passive are that the Indonesian construction is used much more commonly, and that it often cannot be translated by the passive voice in English. The second of these reasons is irrelevant from a linguistic point of view; the first is intriguing and there is a valuable discourse-oriented literature which examines these issues (for example Cumming 1991, Hopper 1988, Kaswanti Purwo 1988 and McCune 1979). However, there are language-internal reasons why the construction should be more common in Indonesian. Extraction (relativisation and question formation) is largely restricted to subjects in Indonesian, not including adjuncts, apart from some exceptions which will be discussed later (and see also chapter 5). Therefore, the di-V construction is very commonly used to feed extraction and might be expected to be more common. The data collected by E. Anderson (1983) suggests that politeness and social distance are also relevant considerations. Discourse studies are not always as exact as might be desired in specifying whether such factors are controlled in their figures, but it seems likely that in narrative at least, there is still a tendency for the di- construction to be used in non-initial clauses of event sequences26. There are more serious syntactic reasons to doubt that all the diconstructions are passive. I assume here that passive is a lexical process which derives an intransitive verb from a transitive verb, the subject of the new verb being thematically equivalent to the undergoer of the transitive verb, and the lexical process being formally marked on the intransitive verb. The diconstruction does not clearly fulfil either of these criteria. The prefix on the verb changes from meN- to di-, but it would be perverse to describe either of these as formally unmarked in relation to the other. Unmarked verbs occur in Indonesian (see examples 4, 5 and 6 above), and both men-V and di-V constructions are formally marked relative to them. But by another, and in this case more compelling criterion, the meN-V construction must be considered as the basic construction of the language: a correct sentence with this verb form can be given for any pair of arguments and any register. Both the di-V construction and the Pro-V construction seen in example 4 impose restrictions on the actor argument, and the bare verb construction seen in example 6 is not useable in formal Indonesian. meN-V clauses are distributionally unmarked, and must be taken to be the basic clause type of the language. The confusion of levels implicit in this terminology is endemic to the literature on Indonesian (SM). 25 Helen McKay (p.c.) reports that in a corpus of contemporary Indonesian journalism, the diconstruction is used with a frequency comparable to English passive. However, this genre is influenced by English in both vocabulary and structure. 26 63 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 In terms of syntactic transitivity, men-V and di-V constructions do not contrast clearly either. I take syntactic transitivity to mean the ability of a verb to license two direct arguments; in this case either type of verb apparently meets the criterion. With men- verbs, the postverbal argument always occurs without a preposition, and with di- verbs, the post-verbal A only needs a preposition when it is not adjacent to the verb: 77. Dia 3SG menjemput saya meN.meet 1SG 'S/he met me.' 78. Saya dijemput di.meet 1SG 'I was met by Ali.' 79. Kami dibawa ke bioskop oleh 1PL.EXCL di.take to cinema by 'We were taken to the cinema by Father. ' (IRG:247,248,259) Ali Ali ayah father But the arguments of A&M, introduced above (section 2.2.2) show that this data is deceptive; the bare DPs in examples such as 78 are not in fact direct arguments. Only the attached pronoun -nya is a term in di-V clauses. di- verbs can license two direct arguments, men- verbs always do27; there is an asymmetry between the two, but a weak one. Transitive verbs are sometimes defined as those which can be passivized. On this criterion, the Indonesian pattern is still unclear; given the symmetry between the two constructions just discussed, it is not impossible to consider men- verbs as the passive (or rather antipassive) of di- verbs. In other Austronesian languages such as Balinese and Tagalog, the undergoer subject clause type has the formally unmarked verb, and an ergative-antipassive analysis has been proposed for such languages by various scholars (Artawa 1999, Blake 1988, De Guzman 1988). Various accounts have been offered in the literature as to how the actor argument of a di-V construction should be analysed, and I now discuss three of these before presenting my own conclusions. 2.3.1.1 THE GENITIVE AGENT HYPOTHESIS All DPs in Tagalog and a number of other Philippine languages have an obligatory case marker or determiner. The marker used for non-subject direct arguments ng [νΑΝ] is identical, or at least homophonous, with the marker used for dependents within DPs: Excepting of course the small number of intransitive men- verbs such as menangis 'weep, cry'. 27 64 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 80. iniabot ng manggamot sa sundalo soldier UF.REAL.within.reach GEN IRR.AF.medicine OBL ang itlog egg NOM 'The physician handed the egg to the soldier.' (Bloomfield 1917: 30) 81. bata' ng dalaga child young woman GEN 'child of the young woman' (Himmelmann 1991: 8) The glossing practice in the above examples is quite commonly followed by Philippinists, with the marker uniformly glossed as GEN(itive). Whether this common marking should be taken to indicate a common function is the subject of some debate. Two questions must be distinguished here: (i) are all non-subject arguments treated the same regardless of thematic role; and (ii) are (some) dependents of verbs functionally identical to dependents of nouns. The first question can be answered definitely only for Malagasy, where the two types of dependent relations are distinguished morphologically for both non-subject clausal arguments and DP-internal arguments (Mahdi 1988). In Tagalog, there is no formal difference amongst non-subject clausal arguments; the actor of an undergoer subject clause and the undergoer of an actor subject clause are both marked with ng, and when an oblique argument is made subject, both actor and undergoer are marked by ng. Some researchers have reported differences in syntactic behaviour dependent on thematic role (e.g. De Guzman 1997), but the data are hard to interpret, given the effects of thematic prominence in Tagalog grammar noted by Schachter (1976). And the most detailed study of Tagalog syntax published in recent years (Kroeger 1993: 47) concludes that non-subject agents are direct arguments (terms) just as non-subject undergoers are. The second question addresses the issue of whether the same functional relation can obtain between a verb and its dependent and a noun and its dependent. The formal identity of case marking could indicate that the functional relation between the two elements is the same in each case, or it could be only a case of available formal resources being deployed in two distinct contexts where there is no possibility of ambiguity. It should be noted that the issue here is distinct from the problem posed by such syncretisms as that of ergative and instrumental case in many Australian languages (Blake 1994: 25-26) where two clause-level functions are involved. If the two functions are taken to be identical, this would lead to a view of Tagalog clause structure and DP structure as very similar, if not identical. Naylor (1980) has argued that this is indeed true for Tagalog, that ng simply marks that an attributive relation of some sort holds between the head and its dependent. She carries these views on to their logical conclusion in a later paper (Naylor 1995) which argues that all Tagalog clauses are equational predications which specify the identity between two nominal elements. On this view, all nonsubject arguments are within DPs, and the case marking reflects this. Voskuil (1996, Ch.7) attempts to construct a theory of case for Western Austronesian languages which assumes that there is a clause-level genitive 65 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 case, and that this is distinct from a clause-level accusative case. Genitive is the case of non-subject actors on this analysis, which allows Voskuil to maintain that undergoer subject clauses are passive and have a different case array from active clauses. This analysis is dubious on several grounds. Firstly, the distinction between genitive and accusative case is formally justified only in Malagasy. In the other languages Voskuil considers, Tagalog and Indonesian, the distinction does not exist in form, and he gives no evidence and no arguments to support the distinction in these languages. Secondly, the evidence presented by Kroeger (1993) for Tagalog, and that presented by Arka and Manning (to appear) as well as the data presented below for Indonesian, all go to show that (some) non-subject agents in these languages are direct arguments. Thirdly, Voskuil himself builds extensive argumentation on the fact that the non-subject argument in meN-V and di-V clauses have identical properties. This is surprising if the two are different types of arguments, appearing in different though formally indistinguishable cases. Finally, Voskuil claims that there are parallel case marking possibilities and mechanisms in clauses and DPs, but he does not allow that the similarity might be deeper than that, he disavows an analysis in terms of nominalisation as discussed above. But his position is not logically satisfactory in this respect. For all these reasons, I cannot accept Voskuil's analysis. 2.3.1.2 AGENT INCORPORATION Another response to these problems is to claim that the diconstructions are not a unitary phenomenon. One study along these lines is Myhill (1988), who argues that when the actor argument is introduced by a preposition, the di-V-PP construction is a straightforward passive, but that when the actor argument is a postverbal DP (di-V-DP), it is incorporated into the verb. The evidence supporting this claim is that the bare DP must be adjacent to the verb, and that it tends to be non-specific. There are several weaknesses in Myhill's proposal. The most immediate one is that a postverbal actor need not be a single word. This must cast doubt on the notion that such nouns are incorporated as, across languages, consistently only single nouns are incorporated, as acknowledged by Myhill, citing Mithun (1984). Myhill suggests two responses to the problem: either the entire DP is incorporated, or only the head noun is incorporated (p120). In defence of the first possibility, Myhill claims that 'prototypical noun incorporation involves only one word', but that this should not rule out the possibility of whole phrases being incorporated. But he presents no evidence in support of this position28, and the source on which he relies for the claim of prototypicality (Mithun 1984) does not make any such claim on my reading, and indeed has no examples of more than one word being incorporated. If the alternative Myhill's statistical evidence of factors which condition the use of the construction is open to essentially the same criticism. He examines a number of factors which predict the occurrence of the construction, none of which is categorical. This is as expected in one direction, i.e. not every clause which meets the conditions for incorporation will actually use it, but is highly surprising in the opposite direction, i.e. not every clause which exhibits incorporation meets the conditions. 28 66 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 analysis is correct, there should exist some syntactic evidence that the remnant of the DP is a separate constituent; Myhill presents no such evidence. Mithun (1984) argues that an incorporated noun is non-referential, and that therefore 'it is not marked for definiteness or number' (849). Myhill's own examples show that the actors he claims are incorporated do not meet this criterion: 82. Api yang kuning menyala itulah yang dipandang fire yellow meN.burn that.PRT di.look.at REL REL keduanya COLL.two.3 'The burning yellow fire was what was being looked at by the two of them.' (Myhill 1988: ex18) In this example, the actor argument is quantified: there are exactly two people referred to. The argument is also specific, with the enclitic -nya used to mark definiteness (see section 1.2.3). Indonesian bare DP actors clearly also fail this test for incorporation. Further problems emerge with Myhill's use of Mithun's 1984 study as support for his position. The part of Mithun's work which Myhill sees as particularly relevant to the Indonesian case discusses the use of incorporation to systematically background known information. There is little doubt that diV-DP clauses in Indonesian can be used for this purpose, even more so when the actor argument is the clitic pronoun -nya, but Mithun sees this as the typical manifestation of type 3 incorporation in her scheme. This is explicitly stated to occur typically in polysynthetic languages, and to imply the existence of type 1 and type 2 incorporation in the language also. Indonesian is not a polysynthetic language and has little trace of type 1 incorporation; there are a few lexicalised verb plus noun combinations, but the process is not productive and such verbs almost always occur without any prefix (Verhaar 1984a:48-9)29. Type 2 incorporation, in which the argument structure of the clause is modified, does not exist in Indonesian. Even if the construction Myhill analyses as incorporation is treated as type 1, the combination of verb and noun as an intransitive predicate, expected properties are lacking, for example there is no tendency for common combinations to be lexicalised. This is certainly related to another problem in Myhill's account: the implausibility of agent incorporation. Just as verb-agent idioms are much less common than verb-patient idioms, lexicalised agent plus verb units are unlikely to occur because we conceive of agents as more independent from actions than patients (Dowty 1991). Again, although Myhill claims that there is no theoretical impossibility in the analysis, it clashes with an implicational hierarchy established by Mithun which has patient as the most likely semantic role to be incorporated. Indonesian has only limited evidence of patient incorporation, and even less sign of incorporation of other roles. Myhill's analysis thus suggests either that Mithun's hierarchy is wrong, or Indonesian does not follow what is otherwise a well-supported universal. A See section 5.3.5 for discussion of another clause type that has been claimed to involve incorporation. 29 67 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 possible escape would be to claim that incorporation works on a syntactic basis rather than a semantic basis, and that objects are the primary target. Then Indonesian di- verbs might be analysed as ergative constructions with the agent as object, and the incorporation would be as expected. However, this involves two additional assumptions (that di- verbs with DP actors occur in ergative clauses, and that the agent of an ergative construction is an object [see Manning 1996a for discussion]) and contradicts an explicit statement of Mithun's, that her generalisations are independent of the syntactic systems of different languages. For these reasons, I reject Myhill's proposal30. 2.3.1.3 TERMS AND NON-TERMS Arka and Manning (to appear, afterwards A&M) also argue that di- verb constructions do not form a single group, but they separate the constructions differently. Based on evidence from binding phenomena, A&M claim that di-Vnya clauses should be distinguished from the other two possibilities31. The crucial evidence depends on the question of whether the agent can bind a reflexive in subject position. On a purely configurational account of binding (Chomsky 1981), this should be impossible. However, a theory which allows thematic prominence a role in binding might make a different prediction, and it is well known that in other Western Malayo-Polynesian languages binding is constrained by thematic rather than syntactic prominence (Schachter 1976). The theory advanced by Manning (1996a,b) posits that binding relations are computed on a syntactic level of argument structure, which respects the division between terms and non-terms, with thematic prominence ordering the arguments within these two divisions (see discussion in section 1.3.3). Thus, in Manning's theory, if an agent is a term, it will be able to bind a reflexive in any other argument position (even subject), but if the agent is a non-term, it will not be able to bind any term. The crucial data is as follows: 83. ?*Dirinya tidak diperhatikan Amir di.care Amir self.3 NEG Himself was not taken care of by Amir. (A&M: ex29a) 84. Dirinya tidak diperhatikannya self.3 di.care.3 NEG S/he didn't take care of her/himself (A&M: ex25a) In example 83, the DP agent Amir cannot bind a reflexive in subject position and therefore it is not a term. The argument structure of the clause is: 85. <dirinya : Amir > where items left of the colon are terms and those on the right are not. On the other hand, in example 84 the reflexive in subject position is acceptable and Clynes (1995) makes a claim similar to that of Myhill for Balinese, but for actor subject clauses: 'Indefinite nominal complements of Actor Pivot verbs are incorporated into the verb to form a syntactically intransitive unit' (196). For the reasons discussed in the main text, this is a typologically more plausible claim. 30 31 Myhill 1988 does not deal with the enclitic -nya at all. 68 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 by Manning's theory the agent clitic must be a term. The argument structure in this case is: 86. < -nya, dirinya : > The actor is a term and as the most thematically prominent term is first on the list. Further support for this analysis comes from the fact that in the case of a ditransitive verb, an agent DP cannot bind a reflexive theme argument: 87. Amiri diperlihatkan Ayahk foto dirinyai/*k Amir di.show father photo self.3 Amiri was shown the picture of himselfi/*k by fatherk. (A&M: ex29c) In this example, the grammatical subject is the only possible binder of the reflexive and the argument structure of the clause must be: 88. < Amir, foto dirinya : Ayah > A&M do not offer an example where the agent is introduced by oleh, but they would clearly predict that such an agent would not be able to bind a reflexive in any term position. Additional evidence supports the analysis of A&M. Indonesian permits floating quantifiers, and this test reliably distinguishes between direct arguments, and obliques and adjuncts (terms and non-terms). The quantifiers seluruh and semua 'all' both normally precede the noun which they quantify, but can on occasion be moved to the end of the clause, sometimes with the addition of the suffix -nya. Keenan (1976) points out that if any type of DP allows quantifier float (QF), it will be subjects, and this is true for Indonesian: 89. Pertahanan-tahanan telah kabur seluruhnya flee all.3 prisoners.DUP PERF 'The prisoners all ran away.' (Chung 1976a: n.30) QF is also possible from both the second argument of a meN-V clause, and from the third argument of a ditransitive verb32: 90. Saya mukul anak-anak itu 1SG hit child.DUP that 'I hit all the children yesterday.' kemarin yesterday 91. Saya memberinya hadiah itu 1SG meN.give.3 present that 'I gave her all the presents before.' dulu before semuanya all.3 semua all But QF is not possible from an uncontroversial oblique, that is a DP within a PP: The quantifiers which float can also be shifted to the left of the NP they modify in a separate phenomenon (see Chung 1976b and discussion in section 1.3.3). This process is not sensitive to the grammatical function of the NP. Therefore in examples 90 and 91 and parallel examples an adverbial appears between the leftmost NP and the quantifier in order to ensure that it is QF that has occurred. 32 69 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 92. Orang-orang Sasak datang dengan anak-anaknya man.DUP Sasak come with child.DUP semuanya all.3 'All the Sasak people came with their children.' NOT *'The Sasak people came with all their children.' The generalisation seems to be that quantifiers can float from direct arguments (terms) but not from oblique arguments. On this test, the nonsubject agent of a di-V clause is not a term: 93. *Ali dipukul mereka dulu semua before all Ali di.hit 3PL (For: 'Ali was hit by all of them previously.') This is consistent with A&M's binding evidence. The evidence then shows that some clauses with di- verbs have two term arguments (those with -nya as actor), while the rest have only one term, the subject. It is therefore wrong to call di- a marker of passive voice: some clauses with di- verbs are passive, but others are not. The correct generalisation, as A&M point out, is that di- indicates that the subject of the clause is an undergoer. I now examine some implications of this claim. 2.3.2 Lexically specified linking As is well known, the Philippine languages such as Tagalog allow arguments with a wide variety of semantic roles, even non-selected arguments, to be subject of the clause: 94. Bumili ang AF-buy NOM lalake man ng GEN isda fish ng GEN pera money sa OBL tindahan store 'The man bought fish with money in the store.' 95. 96. Binili ng lalake ang isda ng pera man fish money UF-buy GEN NOM GEN tindahan store 'The man bought the fish with money in the store.' Binilhan ng LF-buy GEN lalake man ng GEN isda fish ng GEN tindahan store 'The man bought fish with money in the store.' 70 pera money sa OBL ang NOM NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 97. Ipinambili ng lalake ng isda ang pera man fish money IF-buy GEN GEN NOM tindahan store 'The man bought fish with the money in the store.' sa OBL Recent analyses (Foley 1998, Kroeger 1998, Sells 1998) in constraint-based formalisms have treated such alternations as the result of derivational morphology. Verbs (or precategorial roots in Foley's analysis) only specify which semantic arguments must be terms in their basic lexical entries. As can be seen above, these arguments have only two possible realizations: either they are subjects and marked with ang, or they are non-subjects and marked with ng (phonologically [ναΝ]). All other nominals are marked with sa except when they are subjects. This type of system lacks two typical phenomena of voice systems. There is no unequivocal demotion of the actor in undergoer subject clause types, and non-terms are not promoted initially to a non-subject function, that is there is no applicative process. Theories of the linking of semantic roles to syntactic argument positions are based on a correspondence between two hierarchies, a semantic one, the thematic hierarchy, and a syntactic one in which subject is considered to the most salient function available. In the unmarked case, the two hierarchies are aligned, that is, the most thematically prominent argument is linked to the subject syntactic function33 (see discussion in section 1.3.3). Passive can be accommodated within this system simply: the most thematically prominent argument is made unavailable for linking to a core syntactic function and the thematically most prominent remaining argument is linked to subject. This assumption is made fully explicit by, for example, Bresnan and Kanerva (1989), who take it as basic to their theory that agents can be subjects or obliques, but never any other syntactic function. But such an account will not work for a language like Tagalog. In all the clause types in which the actor is not subject, it is apparently still a core argument (see Kroeger 1993 for extensive discussion of the status of non-subject actors). That is, it is still available for linking to the subject function, yet it can be passed over. This argument can be extended by considering the clause types in which more peripheral arguments are subject (the examples above by no means exhaust the possibilities in Tagalog). Leaving aside causatives, in which semantic considerations would lead us to expect the added argument to become subject in any case, the crosslinguistically typical pattern for adding an extra argument (applicative) to a verb is that the new argument becomes a non-subject direct argument, which can then become subject via passivization. But in Tagalog, the first possibility does not exist: there are no added non-subject arguments, and the second stage is obligatory: all added arguments are subjects. Voskuil (1996) claims that all applicative processes in Tagalog trigger obligatory passivization, but without providing any explanatory account of why this should be so. Such an account can preserve the generalisations of linking theory, but only by the 33 I ignore the possibility of an ergative system for simplicity. 71 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 addition of unmotivated processes. In contrast, the analyses mentioned previously take the opposite path and accept that conventional linking theories do not apply to Philippine-type languages. Then the simplest account is that each verbal affix in the voice system specifies directly the linking of one semantic role to the subject function. Clearly, Indonesian does not have a full-fledged Philippine-type system (see Thomas 1978 for an attempted analysis in these terms), but the conclusion reached in the previous section, that di- indicates that the undergoer is linked to subject, suggests that the Indonesian system is related. Four arguments can be advanced to support this position. The first has already been made: that di- is not a passive marker, but something else. The second is that there are restrictions on choice of subject in Tagalog (and other Philippine languages) which are dependent on inherent properties of the argument. While the restrictions in Indonesian oppose pronouns to other nominals and third person to the other two, the Tagalog restriction is related to specificity. In general, Philippine subjects are always specific, and specific undergoers must be subject (Adams & Manaster-Ramer 1988). Nevertheless, in both cases the inherent properties of the nominal are important. The third is that the existence of the bare verb, actor subject construction discussed previously raises questions about the function of meN-. A pair of sentences such as the following are equivalent in propositional meaning: 98. 99. Ali akan melihat ibunya Ali FUT meN.see mother.3 Ali akan lihat ibunya Ali FUT see mother.3 'Ali will see his mother.' The linking of arguments to syntactic functions also is identical; Ali is the subject in each case and ibunya is object. Therefore it is not immediately obvious what the verb prefix meN- actually accomplishes in a clause like example 98. It has the syntactic effects discussed above in section 2.2.2, but it seems unlikely that the prefix is used in order to make enclitics possible or to prevent extraction of non-subject arguments. If the prefix is part of a voice system (using the term rather loosely), then general principles of economy of expression (Bresnan 2001a: 91, Chomsky 1991) would suggest that it is unnecessary34. This conclusion is also supported by the fact that the linking of actors to the subject grammatical function is the cross-linguistic default (excluding ergative languages), and we would therefore expect that pattern to be formally unmarked. Indeed, this is the most plausible interpretation of the clause type seen in example 99; the linking of arguments to syntactic functions is the unmarked, default option. The prefix meN- then must carry some additional information. meN- forms a paradigm with di-, and we have argued that di- specifies the linking of the undergoer argument to subject. The much lower frequency of use of verb prefixes in most varieties of spoken Indonesian suggest that economy is driving language change in this case. 34 72 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 Therefore, we can conclude that meN- specifies the opposite linking, of actor to subject35. The other argument that supports the analysis is that the syntax of Indonesian clauses with prefixed verbs is similar to the syntax of Tagalog clauses. The most obvious example of this is the restriction on relativization possibilities. As discussed in section 2.2.2.2, a verb prefix in Indonesian blocks extraction of any argument except the subject. The same is true of Tagalog, only the nominative argument can be extracted from a clause: 100. isdang ibinigay ng lalake sa bata fish.LNK man child UV.PERF.give GEN OBL 'the fish which was given to the child by the man' 101. batang binigyan ng lalake ng child.LNK PERF.give.DV GEN man GEN 'the child who was given fish by the man' 102. *isdang nagbigay ang lalake sa bata fish.LNK man child AV.PERF.give NOM OBL (FOR: 'the fish which the man gave to the child') 103. *isdang binigyan ng lalake ang bata fish.LNK man child PERF.give.DV GEN NOM (FOR: 'the fish which the child was given by the man') isda fish (Examples from Kroeger 1993: 23-24) The descriptive generalisation here is that no nominal marked with ang (or the equivalent suppletive pronoun forms) can appear within the relative clause, and it is intuitively clear that this fact is related to the fact that the verbal morphology uniquely identifies the semantic role associated with that argument. Various proposals have been made as to the nature of the link between these two phenomena (Nakamura 1998, Sells 1998, Voskuil 1996); the issues are taken up in detail in chapter 5 of this work. At this point, I only draw attention to the fact that the correlation exists and that it appears in Indonesian in exactly the place we would predict given the view of the verb prefixes developed here. Indonesian allows extraction of non-subject arguments, but only when a verb prefix is not present. These arguments lead me to propose the following tentative lexical entries for the prefixes meN- and di-: 104. meN- prefix ___Vtr < x, y, ... > | SUBJ I ignore here the fact that there is a small group of intransitive verbs which have the meNprefix. 35 73 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 105. di- prefix ___Vtr < x, y, ... > | SUBJ Actor and undergoer are not concepts which have any formal status in LFG, therefore the linking stipulation introduced by the prefixes is stated in terms of a-structure. Indonesian verbs with three direct arguments always have the thematically least-prominent argument as second object, therefore these entries only need refer to the first two items in the argument array. The representation 105 makes no statement about the status of the actor when diis prefixed to a transitive verb. I take up this issue in the sections 2.4.3.2-4. 2.4 Pro-V clauses Before presenting my analysis of the Pro-V clause type (sections 2.4.2 4), I first briefly consider an alternative proposal which, on the basis of data from Bahasa Melayu (Malaysian) treats this clause type as having a pragmatic function TOPIC which is distinct from the grammatical function SUBJ. 2.4.1 TOPIC and SUBJECT as distinct functions An analysis of the Pro-V construction is offered by Alsagoff (1991, 1992). On this account, the construction is neither a passive, nor an ergative construction, but rather one in which the pragmatic function TOPIC and the grammatical function SUBJ are assigned to different arguments. Alsagoff denies that the undergoer DP is the subject, treating the actor as such instead. The crucial data supporting this account comes from Malay (essentially the same language as Indonesian). Some speakers of Malay will not allow a Pro-V construction as the embedded clause in an obligatory control construction. Cross-linguistically, it is true that the controlled position is always the grammatical subject in such structures, and this is also true for meN- and diverbs: 106. Ali menyuruh Mariam [_ memukul Ali meN.ask Mariam [_ meN.beat 'Ali asked Mariam to beat the doctor' doktor doctor 107. Ali menyuruh doktor itu [_ dipukul Ali meN.ask doctor that [_ di.beat 'Ali asked the doctor to be beaten by Mariam' (Alsagoff 1991: ex3a, 4a) oleh by itu] that] Mariam] Mariam] Both of these sentences are grammatical for all speakers. But for some Malay speakers a Pro-V structure is impossible in the lower clause: 108. Ali menyuruh doktor itu [_ saya periksa] examine] Ali meN.ask doctor that [_ 1SG 'Ali asked the doctor to be examined by me' (Alsagoff 1991: ex8a) For speakers of what Alsagoff (1991) calls Dialect A, this sentence is fully grammatical, but for speakers of Dialect B, it is not possible. For both dialects, 74 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 there is no reading in which the doctor is the agent of the examining. Alsagoff argues that one of the arguments of the embedded clause must be subject and there must be some reason why it cannot be controlled. The answer to this problem, Alsagoff proposes, is that the clause-initial DP in Malay (and Indonesian) is always TOPIC, prefixed verbs select one or other argument of a transitive verb as both subject and TOPIC, but the Pro-V construction dissociates the two: the patient in clause-initial position is TOPIC (and OBJ) and the immediately preverbal pronoun is subject. Speakers of the restrictive dialect only allow control of positions which are both TOPIC and subject, whilst speakers Dialect A allow any TOPIC to be controlled. The analysis of clauseinitial DPs as TOPIC is motivated by the fact that only this DP cannot be questioned using a wh-in situ strategy, the addition of the clitic -kah to the questioned element: 109.a b 110. *Mariamkah Mariam.kah memukul meN.beat doktor doctor itu that tadi? just now Mariam memukul doktor itukah tadi? Mariam meN.beat doctor that.kah just now 'Was it the doctor that Mariam beat just now?' Mariam memukul doktor itu tadikah? Mariam meN.beat doctor that just now.kah 'Was it just now that Mariam beat the doctor?" (Alsagoff 1991:10a-c) Alsagoff's argument is that questioned elements are focussed, FOCUS and TOPIC are incompatible, therefore the inability of the initial DP to be questioned in this way shows that it is TOPIC. There are a number of problems with this analysis. Firstly, and as discussed in chapter 1, the majority of evidence seems to favour the view that the immediately preverbal pronoun is not grammatical subject. Chung (1976b) reviews a large body of evidence in detail (e.g. control of adverbial clauses), and concludes that the undergoer DP is subject in this construction, and most subsequent studies have agreed with this conclusion. I would place particular weight on one configurational fact: in all other clause types, the subject precedes any temporal or modal operator: 111. Ali sudah membaca meN.read Ali PERF 'Ali has read the book.' buku book 112. Buku itu sudah dibaca book that PERF di.read 'The book has been read by Ali.' 113. Saya sudah baca 1SG read PERF 'I have read the book.' buku book itu that Ali Ali itu that 75 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 114. Buku itu sudah saya book that PERF 1SG 'The book, I have read.' baca read The patient in the Pro-V construction (example 114) is in the same position as the subject of each of the other three constructions, while the agent is in a different position. The patient of a Pro-V construction can be right-dislocated for discourse effect without affecting the relative position of agent and temporal/modal operators36: 115. Bisa kami terbangkan layangan able 1PL.EXCL fly kite 'We can fly the kite' (Chung 1978a:16b) itu that Such examples provide a clear contrast with the bare verb clause type seen in example 113. In a footnote (1992: 163,n51), Alsagoff claims that there is no restriction as to the relative ordering of auxiliaries and actors in Pro-V clauses. This statement contradicts many published sources (McDonald & Dardjowidjojo 1967: 237, Moeliono & Dardjowidjojo ed. 1988: 280-281, IRG: 249, Voskuil 1996: 59-60) and native speaker judgments, and suggests that Alsagoff does not differentiate between the Pro-V and bare verb clause types in her data37. In order to generate the correct word order, Alsagoff proposes phrase structure rules which make the specifier of IP the topic position, with S as the complement of I (1992: 133-134). The position of SUBJ is then immediate daughter of S. These rules are not descriptively adequate. Firstly, they do not account for the possibility of a Pro-V clause without an overt I; in such a case there are two DPs to the left of the verb and there is no obvious structural analysis which allows these two positions (see detailed discussion in section 2.2.1.2). Secondly, if the ordering of actor and auxiliary is not crucial, as claimed by Alsagoff, then it is predicted that two DPs, the TOPIC and the SUBJ, should be able to appear to the left of an auxiliary. Again, the phrase structure rules as presented do not allow this possibility. Such clauses are possible, but they differ in their intonational properties from Pro-V clauses. Thus the Pro-V clause example 116 is spoken as a single intonational unit, but example 117 requires two intonation units with a break after the initial DP38: In the absence of temporals, modals or negation, Pro-V constructions with postposed subjects and bare verb constructions are indistinguishable. Disambiguation is primarily by sentence stress patterns; see Chung 1978a for details. 36 Example 5b from Alsagoff (1991), Saya periksa doktor itu, is starred, but is acceptable to Indonesian speakers subject to the qualifications in the discussion of example 5 above. Whether this is a difference between Bahasa Indonesia and Bahasa Melayu, or a weakness in Alsagoff's data I have not been able to establish yet. 37 The initial DP of example 117 is not a TOPIC by Alsagoff’s test: it can be questioned in situ with –kah. 38 76 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 116. 117. Buku book itu that sudah PERF Buku itu saya book that 1SG ‘The book, I read.’ saya 1SG baca read sudah baca read PAST This evidence is consistent with treating a structure such as example 117 as a left-dislocation structure and therefore distinct from the Pro-V clause of 116. There is an additional empirical argument against Alsagoff's analysis of Pro-V clauses. She treats the actor as subject in such clauses, but this would then contradict a generalisation about subjects in Indonesian: that subjects cannot be reduced pronouns. As exemplified several times (e.g. example 5), the actor in a Pro-V clause can be one of the reduced pronouns ku- and kau-. The undergoer in this clause type cannot be a reduced pronoun, and shares this property with the subject argument of the other three transitive clause types, not to mention all other clause types in the language (this issue is discussed further in section 5.1.2). It might be possible to rescue Alsagoff’s position by restating the generalisation in terms of TOPIC: TOPIC cannot be a reduced pronoun. But this formulation has lesser empirical coverage than the alternative. In example 116, the reduced pronoun ku- can replace saya, but this substitution is not possible in example 117. In conjunction with Alsagoff’s overall account, the proposed constraint makes no prediction about this. In both cases, saya is not TOPIC, and can potentially be reduced. An additional statement about linear adjacency must be included in the grammar to account for the contrast. But the formulation in terms of SUBJ, with the assumption that these clauses are distinct structures, gives the correct prediction without the additional stipulation. In one case, saya is not a SUBJ (example 116) and reduction is possible, while in the other case it is a SUBJ and cannot be reduced (example 117). Alsagoff’s account could be further modified to recognise the bare verb clause type as distinct from the Pro-V clause type, but this would mean abandoning a central claim of her theory, that the selection of TOPICs is accomplished by the verbal prefixes. For the reasons just discussed, Alsagoff's analysis can be rejected for Indonesian, at least. The speakers who allow control into Pro-V clauses accept examples in which the subject (Chung's not Alsagoff's) position in the embedded clause is controlled, as expected. The question raised by Alsagoff's data from the second dialect is not whether this is a subject position, but why some speakers do not allow it to be controlled. I have no explanation for this phenomenon. 2.4.2 Syntax and morphology in Pro-V clauses Salient aspects of the previous discussion can be summarised in a set of descriptive generalisations. In undergoer subject clauses: 1) only pronoun actors can be non-subject terms 2) no pronoun actor (except mereka) can be a non-subject non-term 3) something occurs immediately before the verb in all undergoer-subject clauses 77 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 4) only -nya can express a non-subject actor term after the verb The two items which require specific reference in these statements, mereka and -nya, can reasonably be taken to be special cases. Therefore, my strategy will be to idealise the data and ignore the inconsistencies introduced by these elements. After proposing an analysis of the regularised system which results, I will examine the additional assumptions which are necessary to account for the irregularities also. Thus, the system to be analysed can be characterised by the following three statements. In undergoer subject clauses: 1) only pronoun actors can be non-subject terms 2) no pronoun actor can be a non-subject non-term 3) something occurs immediately before the verb in all undergoer-subject clauses My first step will be to briefly recapitulate and discuss the crucial evidence for each of these generalizations. 2.4.3 Pronoun actors and termhood 2.4.3.1 ONLY PRONOUN ACTORS CAN BE NON-SUBJECT TERMS The crucial evidence in this case is the binding patterns analysed by Arka and Manning (to appear). An undergoer subject can be an anaphor, some version of the reflexive diri, anteceded by the actor in any Pro-V clause: 118. Diri saya saya serahkan ke polisi 1SG surrender to police self 1SG 'I surrendered myself to the police.' (A&M ex.16a) and in a di-V clause when the actor is represented as the post-verbal attached pronoun -nya (example 84). A post-verbal actor which is an DP or within a PP cannot antecede a reflexive in subject position (example 83). If we accept Arka and Manning's version of binding theory which is well-motivated in various publications (Manning 1996b, Manning and Sag 1999, Wechsler and Arka 1998), then the best interpretation of these facts is that, in undergoer subject clauses, the actor is a term when it is a pronoun (ignoring post-verbal mereka) but not otherwise. 2.4.3.2 PRONOUN ACTORS CANNOT BE NON-SUBJECT NON-TERMS Leaving aside mereka, the test case for this statement is the 3rd person singular pronoun dia. This pronoun is fully grammatical as actor in a Pro-V clause, indeed this possibility was one of the factors leading to the hypothesis that di- was also a pronominal element. Typical statements of prescriptive rules for Indonesian grammar claim that di-V clauses are possible with any 3rd person actor, therefore we would predict that a clause such as the following would be possible: 119. ?Buku itu dibaca dia book that di.read 3SG 'The book was read by him/her.' 78 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 But I cannot recall ever seeing a clause like this in written Indonesian, and some speakers reject it outright as ungrammatical. One speaker, however, offered a particularly interesting comment on this example. He said that he could imagine uttering such a clause in a context in which the actor was very contrastive, but that in such a context he could imagine also using a 1st or 2nd person actor as a post-verbal actor. As noted previously, 1st and 2nd person actors do appear as non-terms in prepositional phrases headed by oleh in casual speech, and this possibility is acknowledged in some grammar books. But the possibility of a 1st or 2nd person pronoun directly following a diprefixed verb would not be countenanced. Even in the cases in which a pronominal actor occurs with oleh, the actor is contrastive. In the words of Poedjosoedarmo (1986): If the speaker wishes to put information focus on the agent of the passive verb, the agent must be moved to a position where the rules [of information structure SM] allow it to receive focus (p13) Poedjosoedarmo gives the following example: 120. Itu dibuat oleh saya that di.do by 1SG 'That was done by me.' (Poedjosoedarmo 1986: 13) Therefore, it would seem that this generalisation should be amended to refer only to topical pronoun actors. Pronouns are topical by default, presumably, therefore the cases which limit the generalisation in this way are few and far between, but they do exist. 2.4.3.3 COMBINING THESE TWO GENERALISATIONS The combined effect of the two descriptive generalisations just discussed reduces to a single constraint: 121. (Topical) Pronoun actors must be terms. This constraint can be viewed in several ways. It might be considered to be a statement specific to the grammar of Indonesian which is typologically strange, and an historical explanation might be proposed (see Himmelmann 1996, van den Berg 1996 for studies of Sulawesi languages along these lines). One alternative is to view the constraint as part of an optimality-theoretic system (Prince & Smolensky 1993), that is as a defeasible constraint which is part of the grammar of all languages but whose effects are more or less apparent depending on how it is ranked with respect to other constraints and how it interacts with them39. This constraint would always be satisfied by actor subject clauses, and would be satisfied in all clauses in languages with ergative organization, if we accept that such languages have truly transitive clauses (see Manning 1996a for discussion). The one type of clause which would breach the constraint is the undergoer-subject clause of a language with nominative-accusative orientation, that is the passive. If we assume that The formulation of the constraint in OT terms would be opposite in polarity: *PRO actor NON-TERM - incur a mark for any pronoun actor which is not a term. 39 79 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 the standard view of passive is correct and that this can be reconstructed in terms of OT (see Aissen 1999, Legendre, Wilson and Smolensky 1993, and Sells, to appear, for discussion), then no syntactic means exists for the constraint to be satisfied. The actor is removed from the list of terms associated with a verb by the lexical process which derives passive verbs so it is never available for syntactic expression as a term. However, this leaves open the possibility that a language might have a morphological means of expression for pronoun actors. Transparently, morphology precedes mapping of arguments to syntactic functions (LMT or whatever other theory we assume); lexical processes with morphological consequences like applicative and passive affect the mapping. Therefore in principle a language might be able to satisfy constraint 121 without violating mapping theory. I will argue below that Indonesian is such a language, and I will also argue that once this possibility is allowed, the constraint on the possible means of expressing pronoun actors turns out to be artefactual. The apparent constraint arises as a result of the interaction of the morphological resources of the language with the general principles of Economy of Expression. 2.4.4 Morphology of undergoer subject clauses 2.4.4.1 A PREVERBAL MORPHOLOGICAL SLOT The prefix di- is phonologically and orthographically part of the verb, always immediately to its left. And the pronoun actor of a Pro-V clause cannot be separated from the verb either. This is seen most clearly from the interaction of the restriction on possible actors in the Pro-V clause type and the relative ordering of auxiliaries and preverbal nominals. As discussed previously (section 2.2.1.2), a proper name can be used as a pronoun substitute in a Pro-V clause, but a 3rd person reading is impossible in that context. Thus in 122, where the proper name appears between the auxiliary and the main verb, it can only be read as a pronoun substitute: 122. Mobil saya akan Ali beli Ali buy car 1SG FUT 'You will buy my car.' (addressed to Ali) Where the order of the proper name and the auxiliary is reversed, the natural reading is to give the name 3rd person reference; the pronoun substitute reading is marginal: 123. Mobil saya Ali akan beli Ali FUT buy car 1SG 'My car Ali will buy.' (possibly: 'My car you will buy.' [addressed to Ali]) Also, if an auxiliary intervenes between pronoun and main verb, the pronoun has subject properties, such as being a controlled function in a purpose clause, but this is not true for the pronoun in a Pro-V clause. As topicalization of the object of a bare verb clause is not allowed in a subordinate clause, the word order of the two structures is different, but the contrast is evident nevertheless: 80 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 124. Dia datang ke Indonesia untuk dapat 3SG come to Indonesia for able 'She came to Indonesia to be able to see me.' lihat see 125. Dia datang ke Indonesia untuk dapat saya 3SG come to Indonesia for able 1SG 'She came to Indonesia to be able to be seen by me.' saya 1SG lihat see In both examples, the controlled subject would appear before dapat, but the difference in word order results in a reversal of semantic roles of the controlled subject. When the pronoun immediately precedes the verb, the undergoer is subject, confirming the generalisation stated above. This generalisation is again close to the insight which persuaded earlier scholars that di- must be pronominal. The arguments presented in section 2.2 show that this is not the case, but the arguments presented there also showed that there is no natural analysis of the phrase structure position of the actor pronoun in Pro-V clauses. Therefore, we are forced to examine the possibility that the combination of pronoun and verb is morphological. In fact, ex hypothesi, we should already embrace this position. The existence of the bare verb, actor subject clause type eliminates the possibility that the verb in Pro-V clauses is bare in the same sense, as it is impossible that the same verb form should allow linking of either argument to the subject function. Syntactic processes cannot access the linking of arguments to grammatical functions, therefore some morphology must mediate the linking in Pro-V clauses and the pronoun is the only possible candidate. Two problems arise immediately. Firstly, structures with both a pronoun (or proclitic) and a verb prefix are totally ungrammatical: 126. *Akan saya melihat 1SG meN.see (FOR: 'I will see Ali.') FUT 127. Ali Ali *Ali akan saya dilihat Ali 1SG di.see FUT (FOR: 'Ali will be seen by me.') Assuming that the pronoun and verb combination is morphological, and that information about the mapping of arguments to syntactic functions is conveyed by the preverbal material in meN-V, di-V and Pro-V clauses40, this fact might be taken as showing that there is a single slot which can be filled by elements from two different paradigms. This would be an awkward analysis, but it is not hard to show that it need not be correct. It must be true that both verb prefixes and immediately preverbal pronouns contribute information about linking of arguments, because when neither is present a default linking is adopted. If preverbal pronouns (or perhaps the morphological position they occupy) carry the information that the undergoer Of course, in Pro-V clauses, information about the semantic content of an argument is also present. 40 81 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 is linked to the subject function, then it is unsurprising that the prefix meNand a preverbal pronoun cannot co-occur, because they would carry conflicting information. The co-occurrence of di- and a pronoun would not involve a conflict of information, but it would mean that the same information was given twice, and it is reasonable to assume that this possibility is therefore ruled out on grounds of economy. At this point, it is possible to see that Indonesian undergoer subject clauses do show a split dependent on the semantic properties of arguments, that is a animacy hierarchy split of the type described by Silverstein (1976). Pronoun actors must be terms (taking mereka as a special case) and nonpronoun actors cannot be terms. There is an overlay of discourse-functional effects which cloud the issue, but essentially it is the semantics of the actor which drive this split. But this semantic split does not coincide with the split between the two clause types, because di-V-nya clauses have a pronominal actor which is a term. The formal split between the two clause types is governed by the morphological resources of the language, not the nature of the actor argument. Traditional formulations which describe the two clause types as overlapping in the case of 3rd person actors are not fine-grained enough. The second and more serious problem is that I have argued previously that pronoun actors in the Pro-V clause type are not affixes (section 2.2.1.2). Therefore, the analysis will depend on developing a view of morphology in which that module of grammar is not limited to word-formation processes. I turn to this question in the next section. 2.4.4.2 THE LIMITS OF MORPHOLOGY In previous sections, I have argued that there is no obvious syntactic analysis of the Pro-V construction and section 2.2.1.2 argued that the evidence for treating even the reduced pronouns in the Pro-V clause type as affixes was not convincing. Therefore a morphological analysis would seem at first to be also out of the question. However, two recent studies (Poser 1992, Sadler 1998) have presented strong arguments that the morphological component of grammar is not identical to the word formation processes (including affixation), but that some processes of combining separate words should be considered as morphological rather than syntactic The notion that a word form derived by a restricted rule will pre-empt the derivation of another form by a more general rule has a long history (see Andrews 1990 and Poser 1992 for references). The idea was formalised by Kiparsky as an instance of the Elsewhere Principle (Kiparsky 1982) and can be precisely stated in a unification grammar framework, as shown by Andrews 1990 (see also Blevins 1995). Poser (1992) develops in detail the insight that a similar effect can obtain between a word form and a phrasal construction, that is, the expression of a morphological category may exhibit a complementary distribution between a synthetic and a periphrastic form41. For example, This insight is also present in previous work. Andrews (1984) already analyses English adjective forms along the same lines as Poser, and Di Sciullo & Williams (1987) also note this case. 41 82 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 English adjectives form comparatives and superlatives in one of two ways, either by affixation of -er and -est, or by combination with more and most. The choice is determined by prosodic factors which are irrelevant here. What is relevant is that the two strategies are in complementary distribution: 128. big small good funny silly childish regal damaging symmetric vivacious (Poser 1992 ex. 24) bigger smaller better funnier sillier *childisher *regaler *damaginger *symmetricer *vivaciouser *more big *more small *more good *more funny *more silly more childish more regal more damaging more symmetric more vivacious Another example discussed at length by Poser is that of the Japanese periphrastic verb construction with suru 'do'. In this construction, the verb is preceded by a noun (often a loan word) to create a complex verb: 129. denwa suru sanpo suru doraibu suru nokku suru (Poser 1992 ex. 1,2) to telephone to take a walk to drive to knock Japanese also has a productive process of forming deverbal nouns: 130. Verb Stem Derived Noun Gloss ir iri parching kari kari borrowing mamor mamori protection oyog oyogi swimming sabak sabaki judgment (Poser 1992 ex. 9) We might then predict that it would be possible to form a deverbal noun, and then to use that noun as part of a periphrastic verb. But in fact, such derivations are systematically blocked: 131. *iri suru parch *mamori suru protect *oyogi suru swim *sabaki suru judge (Poser 1992 ex. 13) Poser argues that this is a blocking effect: the original verb stem exists and therefore the periphrastic form is not possible. And if this is the case, he further argues, then blocking is not restricted to the word formation processes of the lexicon. He notes that it is necessary to constrain such possibilities in some way, and suggests two limiting factors. Firstly, only morphological categories, categories that are expressed affixally in some 83 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 language, can be expressed by morphological processes. Secondly, the cases he examines42 can all be considered as examples of 'small' categories, that is, a phrasal category which dominates only zero-level categories. So the Japanese periphrastic verbs are of category V°, dominating a N° and another V°. At least in the case of the English adjective data, this second condition is not adequate as comparative measure expressions admit of phrasal expansion (e.g. exactly three times more expensive / exactly three times dearer. See Bresnan 1973, Bresnan 2001b). It should also be noted that a unificationbased account such as that of Andrews (1990) covers all the data discussed by Poser. Andrews’ theory is discussed in more detail in section 2.5.2; see also Bresnan (2001b) for further discussion of blocking in an optimality-theoretic system. The elements immediately in question, those that are used in the Pro-V clause type fall under both of Poser's conditions. Non-subject pronoun arguments can be expressed by affixes in many languages such as Yimas43: 132. pu-ka-tay 3PL.O - 1SG.A - see 'I saw them.' 133. pu-ηa-tay 3PL.A - 1SG.O - see 'They saw me.' (Foley 1991: 196) Therefore this is a morphological category by Poser's first criterion. Poser's second criterion is also satisfied if we accept that GHT are right about the category of the preverbal pronoun in Pro-V clauses. I argued above that their analysis was incorrect as to the prefix di- being of the same category, and also as to the position in which the pronoun is generated. But I see no reason to doubt their claim that the pronoun (or pronoun substitute) is of category Dº. Whatever the category of the pronoun is, it is clear that it must be a zero-level category. It cannot be modified in situ and it cannot even support a floated quantifier: 134. Buku itu mereka baca semua read all book that 3PL POSSIBLY: 'They read all the books.' BUT NOT: 'They all read the book.' The pronoun is a head which does not project, and it must be a head of a different type from nouns in general as they cannot appear in the preverbal position in this clause type (except as pronoun substitutes). The other generally accepted head with nominal characteristics is D and this therefore is the logical choice44. Two analyses of the range of items which can appear 42 Including some Basque data not discussed here. Glossing as in source: A and O are used following Dixon (1979) to identify respectively the more agent-like and more patient-like arguments of a transitive verb. 43 See section 1.3.3 for argument supporting the claim that Indonesian has the functional category D. In the event that the claim is not valid, an alternative might be to claim that pronouns are the only type of nominal which can appear without projecting additional 44 84 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 seem possible: either the lexical item consists of the features associated with the pronominal meaning, and address terms and proper names are contextually determined phonological realizations of those features; or a lexical rule allows any expression which can be used to address an individual to also be a pronoun substitute. Nothing seems to hang on the choice between these two. Sadler (1998) uses arguments similar to those of Poser to support an analysis of certain clitics in Welsh. The salient facts show strong similarities to the Indonesian facts discussed here. Thus for Welsh Sadler claims: A pronominal argument almost always fails to appear solely in canonical argument position, but is (obligatorily) expressed by means of either cliticisation or pronominal incorporation. (p25) This is exactly parallel to the claim developed above, that for the relevant arguments (non-subject pronoun actors), Indonesian does not allow realization in canonical argument position, but only permits a morphological expression. Sadler also notes that a clitic or incorporated pronominal can never be doubled by a full lexical argument, also exactly paralleled by the Indonesian facts45. She shows that there is no good evidence to treat the Welsh clitics as affixes, but that there is no satisfactory phrase-structure analysis for their positioning, again a parallel with the Indonesian facts. In order to explain the properties mentioned above, Sadler analyses the clitics as Dº heads adjoined to the Xº head of their projection. The crucial move made explicit in Sadler's analysis is that the adjoined element in such a structure can express an argument function46. Welsh clitics certainly do express argument functions: object when attached to V, subject when attached to I and possessor when attached to N, and if the lexical adjunction analysis is accepted the remaining questions are how the argument function is assigned and how the possible assignments are constrained. On the first point, Sadler notes that the same clitic forms can express different functions depending on the item they are attached to, the clitics do not have different case forms. This means that argument functions are not associated with the clitic forms in the lexicon, as is the case for dependent-marking languages with varied case forms. The clitics are not agreement markers either, they cannot be doubled by free nominals. Therefore head-marking is not being used to identify argument functions either. The remaining possibility in LFG is that the argument function is assigned configurationally, by functional annotation of a c-structure schema: structure. But this claim would presumably be rooted in the semantics of specificity and would imply the further claim that bare nouns appeared with a null element of some kind which supply the necessary additional semantics. Such an analysis is a notational variant of that which invokes D as a category. Welsh allows a copy pronoun in some cases, but does not share this property even with the closely related Irish Gaelic (Andrews 1990, Sadler 1998: 27) 45 It is not clear to me whether this move is implicit in Poser's analysis of Japanese periphrastic verbs. He calls such structures 'incorporated' which suggests that the light verb is not assigning an argument function to the associated nominal, but this is never stated definitely. 46 85 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 135. Example 135 is to be interpreted as stating that a Do can be adjoined (morphologically) to a lexical head, and be assigned an argument function in the f-structure associated with the head. This leads naturally to the second question: are there constraints on the argument function which can be assigned by any given lexical head? Sadler suggests that only local argument functions can be assigned in this way, local in the sense that the argument's position in c-structure is within the maximal projection of the head. Sadler's precise formulation is: 136. A language may make use of a lexical adjunction structure to express argument functions. The argument functions which can be expressed are limited to those generated within the Xmax projection of the c-head Xº by the endocentric mapping principles. (p25) Bresnan (2001a: 98-109) gives a full discussion of endocentricity; the effect of the principles is to allow the possibilities noted above: a clitic attached to V is an object, one attached to I is a subject and one attached to N is a possessor. Again, the Indonesian data shows striking similarity to the Welsh facts. Indonesian is not a case-marking language. There are two instances in the language of lexical items having different forms depending on their syntactic environment. One is the 3rd person singular pronoun dia/ia discussed briefly in section 2.2.1.4, a relic of a prior stage in which the language had some case-marking. The other is the 2nd person bound pronoun which has the form kau- in the Pro-V clause type and the form -mu in all other positions. In other words, one form occurs only in the case in which a pronoun can be attached to the left of its host, the other form occurs in the other cases where the pronoun attaches to the right of the host. This should be regarded as a phonological phenomenon as there is no consistent relation between form and grammatical function in this case. In the most comparable cases, that of -mu attached to a meN- prefixed verb, and kau- in a Pro-V clause, the attached pronoun is a non-subject in both, and can be treated as OBJ in each also. The correlations which exist then are with semantic role and with linear position. The semantic role of agent (actor) is also assigned by the preposition oleh, and it can host the pronoun -mu but not kau-: 137. Hal itu kedengaran olehmu matter that overhear by.2 'That matter got overheard by you.' 138. *Hal itu kedengaran olehkau Therefore linear position is the conditioning factor, and the difference can be treated as phonological. As noted previously, Indonesian also lacks verbal agreement of any kind. As discussed in section 2.2.1.3, and in this section, doubling of attached pronouns is never possible. The pronouns do not reference arguments, they are arguments. 86 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 Using the work of Poser and Sadler as a basis, an analysis of actor pronouns in Indonesian as morphological elements is very plausible. In the following section, I present such an analysis, beginning with Pro-V clauses. 2.5 An analysis 2.5.1 Pro-V clauses I adopt the following assumptions which have been argued for above: 1) Pronouns are generated as Dº. 2) Adjunction of zero-level categories is a possible morphological operation, at least for Vº and Dº in undergoer subject clauses. Assumption number 2 needs more comment. As discussed above (section 2.4.4.1), the preverbal pronoun codes the linking of undergoer to subject as well as the pronominal meaning. This must be the case, as a bare verb with no morphological dependents has default linking of arguments to grammatical functions i.e. actor is linked to subject. Therefore a morphological rule is required rather than a statement that the adjunction is possible47: 139. Pro-V Adjunction V → Dº actor Vtr ↑ ARGF = ↓ ↑=↓ [PRED 'pro'] < x, y, .... > | SUBJ It is not necessary for this rule to specify what argument function the pronoun will be assigned48, if we allow that the actor in an ergative construction is an object. Sadler's principle (136 above) predicts that only an argument function generated within the maximal projection of the c-head will be possible. The chead is V, and object is the grammatical function canonically generated within its Xmax. Given that the linking for subject is specified, this is not a point of major interest, but it is preferable to use the less stipulative formulation, particularly as this accords with a generalisation which has some validity cross-linguistically49. 47 The case of ditransitives is handled as noted previously: see section 2.3.2. I assume that the actor in an ergative construction can be an OBJ: see Bresnan & Kanerva (1989, n32) and Manning (1996a) for various opinions on the question. 48 It is not clear to me how this principle operates when there is no auxiliary in a clause and V is an extended head occupying I in c-structure (see section 1.2.3). In such a case, the endocentric mapping principles which Sadler (1998) relies on could give two results: if the head is treated as V, the OBJ function will be available, but if it is treated as I, the SUBJ function will be available. I assume that lexical categorisation takes precedence in such a situation. 49 87 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 The effect of this rule is as follows. A transitive verb has a lexical entry such as the following (I use only actor and undergoer as shorthand for the semantic roles assigned by the verb): 140. lihat Vtr 'see' < ACTOR, UNDERGOER > The morphological rule 139 applies, giving rise to the following structure and its associated partial f-structure: 141. 142. This verb can form a part of a larger structure, following general principles, with the following result: 88 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 143. 144. This analysis accounts for various properties of the Pro-V clause type in a natural fashion: 1) Treating the pronoun and verb unit as a morphological combination resolves the issue of the phrase structure position of the pronoun, explains why the two cannot be separated and why the pronoun is a non-projecting category. 2) It was noted previously (section 2.2.2.2) that we might expect the nonsubject argument of Pro-V clauses to be extractable as the verb is not prefixed. Under the present analysis, the possibility of extraction is ruled out by the fact that the verb is not really bare, it has morphological material associated with it; and by the fact that the non-subject argument is realized morphologically and is therefore not visible to syntactic operations. 3) Allowing an argument function to be associated with a morphological element explains the binding theoretic properties of the Pro-V clause type, while also constraining the possibility of non-subject actor terms in accord with the data (see also the following section for binding in di-V-nya clauses), and in accord with general considerations. The data on co-ordination in this type of clause (discussed in section 2.2.1.2) also fit in with the current analysis. It is impossible to co-ordinate two pronouns as actors: 89 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 145. *Anjing itu saya dan dog that 1SG and (FOR: 'I and he hit the dog.') dia 3SG pukul strike This possibility is ruled out on two grounds. Firstly, co-ordination is a part of syntax and therefore is not available as a part of a morphological process, and secondly, for any combination of pronouns there is a single pronoun which could be used instead and the co-ordinate structure is therefore ruled out on grounds of economy of expression. Thus in example 145, saya dan dia 'I and he' could be replaced by kami '1st person plural exclusive'. As for the co-ordination of two verbs with a single actor pronoun, this would seem to be ruled out again by the non-availability of co-ordination in morphological processes. The marginal acceptability of such structures for some speakers can be attributed to the transparency of the intention. 2.5.2 Adding -nya to the picture The qualification which -nya added to the system, noted at the start of section 2.4, was that it is the only actor pronoun which can occur after the verb (modulo the effect of contrast and focus discussed previously). It is natural to associate this difference between -nya and the other pronouns (still excepting mereka) with the fact that it is the only pronoun that all speakers use for non-human reference (previously discussed in section 2.2.1.3). Prescriptively, the 3rd person pronouns dia and mereka have only human reference, although in informal usage this distinction is not always maintained. But such usage is not universally accepted and is seen by most commentators as innovative (see Kaswanti Purwo 1984 and IRG: 167-168). We might then revise the rule for Pro-V adjunction (139 above) to include the constraint that the adjoined pronoun must include the feature [HUMAN +] in its matrix. But the fact that -nya following a di- prefixed verb codes a term suggests that perhaps rule 139 needs more far-reaching revision. The null hypothesis must be that the same morphological process operates in the case of both preverbal and postverbal actor pronouns, that is that the morphological adjunction process does not have ordering specified. Unless there is strong evidence to the contrary, we must adopt this position, which would lead in turn to the necessity of separating rule 139 into two components. One of these would express the generalisation that a personal pronoun adjoined to the left of a transitive verb results in an undergoer subject verb, while the other would express the generalisation that any pronoun adjoined to an undergoer subject verb is assigned an argument function. The evidence shows that -nya following a di- prefixed verb is in a morphological relation to the verb, whereas -nya following a meN- prefixed verb or a preposition is a true clitic in a syntactic relation with its host. The previous discussion has established that the actor in a di-V clause is a term when it is realized as -nya, but not when it is realized as a DP adjacent to the verb. This is a surprising result if -nya is a clitic in the commonly understood sense, that is, a syntactic terminal in its own right which is phonologically 90 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 dependent on another word. If -nya is a clitic in this sense, then we would assume that it would be a constituent of the same category as any other nominal (regardless of whether we label this as NP or DP) and it would be in the same phrase structure position relative to the verb. Therefore, the difference between the term, in the one case, and the non-term, in the other case, would be unexplained. We could stipulate that a pronoun in that position was a term, but even this ad hoc solution is not sufficient, as the pronoun mereka can occur in the position (see example 28), but it is not a term: 146. Sendiri*i/j diperhatikan merekai self di.look.after 3PL 'Theyi look after themselvesj.' Co-reference is not possible in this case; the actor cannot antecede a subject reflexive pronoun. Additional evidence as to the status of -nya in di-V clauses comes from co-ordination. I suggested above (section 2.1.1.2) that co-ordination possibilities did not provide a reliable test to discriminate between clitics and affixes. However, in this case, I believe that the evidence is useful because the judgments seem to be more definite than those for co-ordination in Pro-V clauses, and because the co-ordination data are not the only evidence available. The binding data already establish that -nya in a di-V clause has a different status to other actors in such clauses. But quantifier float data shows that such actors are not terms in the same way that full DP arguments are. Some speakers will allow -nya to have plural reference (see also section 2.1.1.3): 147. Saya sudah membacanya 1SG meN.read.3 PERF 'I've read it/them.' (IRG: 167) But speakers who allow plural reference for the pronoun will not allow a floated quantifier to modify it: 148. ?(*)Buku itu dibacanya dulu semua book that di.read.3 previously all 'All the books were read by him/her/them (for some speakers).' NOT: 'The book was read by all of them.' This is parallel to the data for preverbal actor pronouns discussed in the previous section. Attached pronouns such as -nya occur in three contexts in Indonesian: following prefixed verbs, following some prepositions, and as possessors within DPs. Following a meN- prefixed verb or a preposition50, these pronouns behave like standard clitics, and co-ordination with a following DP is possible: 50 I do not consider possessors here. For further discussion of attached pronouns, see chapter 5 91 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 149. Saya mencintaimu dan 1SG meN.love.APPL.2 and 'I love you and your mother.' ibumu mother.2 150. Saya pergi dengannya dan Ali 1SG go with.3 and Ali 'I went with him and Ali to the village.' ke to kampung. village But such coordination is not possible with -nya following a di- prefixed verb: 151. *Siti dilihatnya dan ibunya Siti di.see.3 and mother.3 (For: 'Siti was seen by her and her mother.' Co-ordination for bare DP actors following di- prefixed verbs is fine, even when the first co-ordinated element is a pronoun: 152. Buku itu dibaca guru dan mahasiswa book that di.read teacher and student 'The book was read by the teacher and by the student.' 153. Buku itu dibaca mereka dan book that di.read 3PL and 'The book was read by them and by us.' kita 1PL.INCL This evidence shows again that -nya attached to a di- prefixed verb is not a clitic and that it has a special status. Again, it is parallel to the data for preverbal actor pronouns and again it supports the argument that these pronouns are zero-level elements morphologically attached to the head verb. An alternative analysis would be that –nya represents two homophonous pronominal forms, one strong and one weak in the sense of Cardinaletti & Starke (1996). The impossibility of co-ordination is a diagnostic for deficient pronouns for Cardinaletti and Starke, and the –nya which attaches to di-prefixed verbs would be deficient by this test, whereas the –nya which appears elsewhere would not. However, pronouns which are not deficient are assumed to be phonologically free in this scheme, and –nya always requires a host on its left: it cannot occur on its own. Such an analysis would also ignore the parallels between the preverbal pronoun of the Pro-V clause type and –nya following a di-prefixed verb. The pronouns of the Pro-V clause are strong by another of Cardinaletti & Starke’s tests, in that they can have only human reference, but the –nya following a di-prefixed verb would be weak. This analysis therefore cannot be accepted51. An additional argument for accepting this proposal is that it allows us to account for the contrast between meN-V clauses and di-V clauses with respect It is possible that there is some distinction between human and non-human reference for the clitic –nya. The relevant examples above (examples 149 and 150) have human referents for the clitic, and co-ordination may fail in similar cases where –nya has non-human reference (for speakers who allow this). I have not checked this point, and in any case it is not clear to me whether the point is relevant given the considerations discussed above. 51 92 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 to topicalization from cleft structures. In section 2.2.2.2, I repeated evidence reported by Voskuil (1996) showing that topicalization of the non-subject argument was possible from within a headless relative clause (example 69 repeated): 69. Surat ini saya yang menulis(nya) letter this 1SG meN.write REL 'As for this letter, it is me who wrote (it).' (Voskuil 1996: 193 ex.32) This structure is possible only when the verb prefix is meN-, with di- it is impossible. Voskuil links the possibility of this structure to the possibility of a resumptive pronoun within the relative clause, claiming that there is a resumptive pronoun in all cases, but that it is not always overt. Regardless of the correctness of Voskuil's account, the basic observation is intuitively correct: the resumptive pronoun and the extraction possibilities are interdependent. In section 2.2.2.2, I suggested that as the attached pronoun was possible in the parallel di-V clause, it was surprising that the topicalization failed. But in light of the theory developed in this section, this apparent paradox is resolved. -nya following a di- prefixed verb is not an attached pronoun, it is a morphological element. Therefore the two cases are not parallel. The pronoun is attached to the verb morphologically, and the resultant lexical structure is opaque to syntactic processes such as topicalization. In light of the discussion immediately above, I propose the following two rules which replace rule 139 above and also generate di-V-nya clauses: 154. Undergoer Subject Verb Morphological Rule 155. Undergoer Subject Verb Pronoun Adjunction Representation 154 is intended to mean that a transitive verb which has a personal pronoun left adjoined to it will have the linking of undergoer to subject specified at the dominating V node, and representation 155 is intended to mean that if a V node which has the specified linking of undergoer to subject dominates an adjoined pronoun, the pronoun is 93 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 assigned an argument function by the verb52. Pro-V clauses receive the same analysis as that given in the previous section, as in that case the two new rules have the same effect together as the single rule 139. The analysis of di-V-nya clauses is now straightforward also. The actor is not a personal pronoun, therefore rule 154 does not apply. But the prefix di- can be joined to the verb, resulting in a verb with the linking of undergoer to subject specified (see 105). The principles forcing a topical pronoun actor to be a term still apply, and rule 155 makes this possible, resulting in the following structures: 156. 157. The structure of a complete clause is: Clearly, there is some redundancy between these two rules and this suggests that a deeper generalisation is being missed. I return to the question in chapter 5. 52 94 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 158. 159. This analysis accounts for the binding facts, the co-ordination facts and the topicalization facts. Note that LMT predicts that the actor should never be an OBJ: it has the feature [-o], which should prevent it ever being mapped to any GF except SUBJ and OBLθ. This prediction may be falsified by languages with ergative-absolutive alignment (see footnote 48 and references cited there). In any case, the analysis presented here for both Pro-V and di-V-nya clauses apparently contradicts this prediction. It is possible, though, to maintain the analysis of verb prefixes proposed here, and LMT, given certain assumptions. LMT must only apply after all morphological processes: this assumption is well-motivated and not problematic. Given this assumption, a di-V-nya verb has both its arguments mapped to GFs when it is morphologically complete, and LMT is irrelevant, no further mapping remains to be accomplished except in the case of ditransitive verbs. I assume that in such cases, the feature [+o] will be assigned to the third argument, as the assignment of features is dependent on the semantic status of arguments. The ditransitive verb has two patient-like arguments, regardless of how these are realized, and features can be assigned accordingly. The mapping to OBJθ follows from the principles of LMT. The case of a di-prefixed verb which does not have -nya attached to it is more complex and the most general account depends on assuming one of the versions of a-structure introduced in section 1.3.1 rather than the other. I have argued above that di- does not mark passive, but its effects are nevertheless similar to passive. Passive blocks the mapping of the most thematically prominent argument to the SUBJ GF; mapping follows afterwards 95 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 by general principles. In the case of an undergoer subject verb, the SUBJ GF cannot map to the most thematically prominent argument because that GF has been assigned lexically. That argument has the feature [-o] by general principles, and as SUBJ is not available, the only compatible GF is OBLθ. This account follows naturally within the version of a-structure presented by Bresnan (2001a: Chapter 14), in which the distinction between term and nonterm is not marked. For this version of a-structure, a ditransitive verb has three arguments which can be realized in various ways depending on the interaction of specific features of lexical items and general principles. However, if the alternative conception of a-structure is assumed, this account will not work. If the number of term arguments is specified in a-structure, then a ditransitive verb with the prefix di- and without –nya as actor has a different valence to the base verb: it has only two terms rather than three. In this case, Bresnan’s version of a-structure allows the more general account and is to be preferred. However, in section 3.3.2.3, I discuss a case where the data seem to favour the other conception of a-structure (that of Manning 1996a). An account incorporating the assumptions just discussed is possible. But in chapter 5, I will suggest an alternative. The lexical entries for the verb prefixes (examples 104 and 105) and the morphological rule for pronoun adjunction (example 155) already implicitly claim that LMT is not relevant for the linking of an argument to the SUBJ function as this mapping is lexically specified. In chapter 5 I will extend the argument and suggest that, in the case of prefixed verbs, LMT may not be relevant for the linking to other functions either. Free pronouns appearing after a di-prefixed verb would be non-terms, and are therefore barred by the constraint 121. But the constraint does not bar the appearance of 1st or 2nd person attached pronouns following a diprefixed verb. Two related questions therefore must be answered: why are forms such as dia lihat and dilihatnya both possible, and why is saya lihat possible, but not dilihatku? Intuitively, the answer to both questions will appeal to some blocking principle. saya lihat and dilihatku would have identical possibilities for the reference of the pronoun, therefore one blocks the other, but dia and -nya do not have identical reference, and both are possible. This intuition can indeed be captured in a precise fashion. The principle of morphological blocking has been stated formally in LFG by Andrews (1990): 160. Suppose the structure S has a preterminal node P occupied by a lexical item l1. and there is another lexical item l2 such that the f-structure determined by the lexical entry of l1 properly subsumes that determined by the lexical entry of l2, and that of l2 subsumes the f-structure associated with P in S (the complete structure, after all unifications have been carried out). Then S is blocked. One f-structure, F, subsumes another f-structure, F', if the information in F is a subset of the information in F', if F' is more specific. Therefore this principle states that a lexical item is blocked from appearing in a structure if there is another lexical item which is more specific, but which does not alter the total information content of the structure. The most natural account of -nya is to 96 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 say that its lexical entry includes the information that it is a pronoun, and that it has the value '3' for the attribute PERSON. This would make it compatible with both singular and plural, and human and non-human antecedents53. However, if this approach is adopted, the principle above (160) predicts that a form such as dia lihat will block dilihatnya. The f-structure associated with dilihatnya would be: 161. di-V clause: dilihatnya while that associated with dia lihat would be: 162. Pro-V clause: dia lihat The structure 161 is less specific than the structure 162: 161 subsumes 162. Therefore dia lihat is a potential blocker of dilihatnya but not vice versa. The f-structure associated with the total structure in which dilihatnya appears will include (minimally) additional information about the SUBJ, but any such structure will not be subsumed by the structure 162 which includes additional information about the OBJ. Therefore, in accordance with Andrews' principle, the structure which includes dilihatnya is not blocked: both structures are possible. Different considerations apply in the case of the 1st and 2nd person pronouns. The f-structures associated with, for example, saya lihat and dilihatku are identical: 163. The Morphological Blocking Principle stipulates that one item may block another only if it is properly subsumed by it; in this case the two items Conservative speakers would have values specified for one or both of these other attributes in their lexical entries. 53 97 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 (trivially) subsume each other, but no relation of proper subsumption exists. Andrews (1990, n.12) suggests that in such a situation free alternation is possible. But that is not the case here; some other factor is at work. I assume that the explanation is to be sought in economy of expression. One item, saya lihat consists of a pronoun and a verb stem, morphologically composed, while the other consist of the same verb stem with an affix and a pronoun, all three morphologically composed. There is no reason not to suppose that economy of expression may be a principle in choosing between alternative morphological expressions of the same material, as well as in choosing between morphological and syntactic expressions of identical material. And economy would clearly prefer saya lihat, the observed result. The possibility of focussed pronouns appearing after di- prefixed verbs also fits plausibly into this analysis. The constraint 121 is waived in the case of a non-topical pronoun, and the addition of the information that the pronoun was focussed would remove the blocking relation between the two possible expressions. 2.5.3 Adding mereka The analysis developed to this point predicts that a personal pronoun actor, except when strongly contrastive, will only appear in one type of undergoer subject clause, the Pro-V type. -nya is the only pronoun which can appear in a di-V clause because it is the only one whose feature matrix is not possible in a Pro-V clause and which is therefore not blocked from being expressed in another way. However, the 3rd person plural pronoun mereka can appear in both clause types, contrary to prediction: 164. 165. Buku-buku book.DUP ini this sudah PERF mereka 3PL Buku-buku ini sudah dibaca book.DUP this PERF di.read 'They have read these books.' baca read mereka 3PL There is no obvious syntactic reason why this should be the case, and the only explanation I can offer is a historical one. mereka was not part of the original Malay pronoun system, but was borrowed from Javanese at some point. Zoetmulder (1982) suggests that the Javanese marika was a demonstrative, also used as a 3rd person pronoun. But Adelaar (1992b: 125) cites two early nineteenth century sources on Malay which both have marika as a noun. It is possible then that the word was borrowed into Malay as a noun which soon came to be used as a pronoun substitute. As a noun, the word would be able to appear as actor in di-V clauses (see following section for discussion), and as a pronoun substitute it would also be able to be used in Pro-V clauses. The nominal meaning was lost at some stage, but the postverbal possibility was left as a relic. This scenario gains some indirect support from the usage of pronoun substitutes in modern Indonesian. Kinship terms are used frequently as pronoun substitutes in Pro-V clauses, as previously discussed: 98 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 166. Buku itu ibu book that mother 'I / you read the book.' baca read The same words are freely available as actors in di-V clauses, like any other noun. But in such cases, pronoun substitutes can still have a 1st or 2nd person reading, despite the general constraint against 1st or 2nd person actors in di-V clauses: 167. Buku itu dibaca book that di.read 'Mother read the book.' OR 'I read the book.' OR 'You read the book.' ibu mother This evidence shows that there is an ambiguity as to the status of pronoun substitutes within the overall system, and this gives some plausibility to the scenario sketched above. Of course, real credibility will depend on detailed examination of the use of mereka in Old Malay texts, work which is yet to be done. 2.5.4 Bare DP actors in di-V clauses A&M argue, and I have adopted this position, that the bare DP actor of a di-V clause is a non-term, that is an oblique argument. The surprising consequence of this position is that an DP which is obligatorily adjacent to a verb is not a direct argument of it. This seems to be contrary to standard views of phrase structure. The actor behaves as though it were within the immediate projection of the verbal head, a position normally reserved for direct arguments, but A&M's analysis of the binding facts, and the quantifier float data discussed above, show that it is not a direct argument. Note also that with (derived) ditransitive verbs in the di- construction, the actor DP occurs inside the second object: 168. Amir membelikan saya Amir meN.buy.APPL 1SG 'Amir bought me a shirt' baju shirt 169. Saya dibelikan Amir baju di.buy.APPL Amir shirt 1SG 'I was bought a shirt by Amir' (A&M:28b) I assume that the theme argument in 168 and 169 (baju) is a VP constituent. Therefore there are two possible interpretations of such structures. Either the theme is a direct argument and the structure has an oblique argument closer to the head of the VP than a direct argument, or the theme is also an oblique argument. The first alternative is problematic within standard theories of phrase structure54, but it must be correct because the QF data rule out the This is especially true for a monostratal framework such as LFG; the surface configuration in such a theory cannot be derived by movement operations. 54 99 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 second alternative. As shown above (example 91), the second object of a ditransitive verb can float a quantifier, and can therefore be considered a direct argument. The account of passive which I am assuming here predicts that the passive of a ditransitive verb will be a transitive verb with both the non-subject arguments of the active remaining direct arguments. This prediction is borne out in Indonesian, with the original second object still able to float a quantifier in the di-V clause: 170. Ali diberi mereka buku itu dulu semua Ali di.give 3PL book that before all 'Ali was given all the books by them before.' (NOT: 'Ali was given the book(s) by all of them before.') This phenomenon is not affected by the status of the agent: 171. Ali diberinya buku itu dulu Ali di.give.3 book that before 'Ali was given all the books previously.' semua all Thus the first possible interpretation noted above must be correct: in such constructions a direct argument, subcategorized for by the verb, is further from the verb than a non-term argument. The relationship between a bare DP actor and its verb is unusual in another respect. The adjacency requirement between the two is absolute; if anything intervenes, then the preposition oleh is required before the actor: 172. Film itu dilihat dulu *(oleh) film that di.see before by 'That film was seen previously by Umar.' Umar Umar The behaviour of the two types of prefixed verbs is different in this respect. With meN- prefixed verbs, there is no alternative way of expressing the undergoer55 and material can intervene between the verb and the object: 173. Saya membeli sekarang 1SG meN.buy now 'I bought the book just now.' buku book itu that These facts show that the actor of a di-V clause is licensed in some fashion only if it is adjacent to the verb (or part of the verb morphologically - see previous section), otherwise the preposition is needed to license the DP. The most satisfactory account of these facts would seem to be that a diprefixed verb gives the possibility for a non-term argument DP to be right adjoined to it when that DP has the semantic role of actor. This proposal gives the right word order result, and also suggests a reason for the rigidity of the word order. It also gives a structure in which the verb and the actor are not in This was not the case in Old Malay, where all transitive verbs could use the preposition akan to introduce their object (Cumming 1991). 55 100 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 a head complement relation, but in which the object of a ditransitive verb is the complement of a head: 174. A similar proposal has been made by Cole and Hermon (2000) to account for word order facts in Toba Batak. In fact, Toba Batak exhibits rigid ordering in both actor subject and undergoer subject constructions, with the non-subject argument appearing between the verb and the subject in each case, and the second object of ditransitives appearing outside the subject 56: 175. Mangalean buku si-John natoari book HON-John yesterday ACT-give 'John gave the book to Mary yesterday.' tu-si-Mary to-HON-Mary 176. Dilean si-John sada buku tu-si-Mary one book to-HON-Mary PASS-give HON-John 'One book was given to Mary by John.' In the words of Cole and Hermon: Both the Direct Object and the Passive Agent appear to be "frozen" to the verb. Not only can no element intervene between either of them and the verb, but neither can undergo any movement rule. The verb prefixes involved in the two languages (Indonesian and Toba Batak) are clearly cognate, and it is therefore unsurprising that other properties of the languages should be similar. Note however that Indonesian only has these properties in the di-V clause type; example 173 above shows that word order in meN-V clauses is less rigid57. At this stage, I am not able to offer a non-stipulative account of this phenomenon. Therefore, I only propose a language-specific rule as follows: 56 Glossing as in Cole and Hermon (2000): ACT - active, PASS - passive, HON - honorific. Besides the word order facts, Cole and Hermon's proposal is intended to account for binding facts in Toba Batak which, like Tagalog and Indonesian, allows non-subject actors to bind anaphors in subject position. The present work assumes that binding relations are not solely dependent on phrase structure configurations, and this aspect of Cole and Hermon's work is not relevant here. Keenan (2000) makes a proposal related to those discussed here, that actor and verb form a constituent in some Malagasy clauses. 57 101 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 177. Actor adjunction (Indonesian) V→ di-V DPactor ↑=↓ ↑ OBLθ = ↓ The data and arguments presented in section 2.5.2 establish that there is a difference in the relation of di- prefixed verbs and following -nya and a following DP. Although I have argued that the actor is adjoined to the verb in each case, the nature of the adjunction is different. In one case, the two elements are joined morphologically, in the other case syntactically, and the different properties of the two types of actors follow from this difference. The possibility of licensing oblique arguments by adjunction to the governing head will be discussed further in section 3.3.3.2, where it will become clear that the possibility is more generally available in Indonesian. This concludes my discussion of di-V clause types. I assume that the type of clause in which the actor occurs in a prepositional phrase is a standard passive clause type, and does not need additional discussion here. However, I will revisit this assumption and other aspects of the syntax of di-V clauses in chapter 5, where I consider an alternative view of the structure of transitive clauses in Indonesian. 2.6 Summary In this chapter, I have presented an analysis of the common clause types for verbs with two direct arguments in Indonesian. The fact that there are four clause types in which such verbs can appear raises various problems which have not been satisfactorily dealt with previously. Here, I have argued that diis a verbal prefix rather than a pronoun and that it is in paradigmatic opposition to the prefix meN-. Not all clauses with the prefix di- are intransitive and it is therefore inappropriate to describe it as a passive marker. Rather, it signals that the undergoer of the clause is linked to the subject grammatical function. By symmetry, the prefix meN- must therefore signal the linking of actor to subject, rather than marking active voice in the sense usually understood. This conclusion is reinforced by the existence of a clause type with a bare verb form in which the actor is subject; this type is the crosslinguistic default linking option, an active voice. Consideration of the two types of undergoer subject clauses forces the conclusion that pronoun actors must be terms in such clauses, except when they are contrastive (focussed), and that the only means by which this expression of arguments can be achieved is morphological. Indonesian therefore has a highly unusual transitive system. It has features (specified linking to subject) which are related to Philippine languages, it has only a limited place for features which are the norm elsewhere (default linking of actor to subject), and it has the possibility of undergoer subject clauses with two direct arguments. Normally, one would describe such clauses as ergative. But as has been argued in detail here, this pattern is only possible because of certain morphological properties of the language, and there is good evidence that the actor arguments in these 'ergative' clauses are invisible to syntactic operations. The clause types in question are not 'morphologically ergative' in the usual sense of that term - there is no distinctive form associated with the 102 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 2 actor of these clauses. But it seems equally inappropriate to describe this as 'syntactic ergativity', when the ergative argument is apparently not syntactically represented. I leave the choice of appropriate terminology to the reader. The central concern of this thesis is to elucidate the status of nonsubject arguments and the manner in which they are licensed. The assumption that the system analysed here is that of transitive verbs in Indonesian carries with it the assumption that the non-subject argument of actor subject clauses is an object, OBJ in LFG terms. However, the data in section 2.2.2 has shown that there are significant differences in the properties of this argument in the two types of actor subject clause: in one case, the object can be an attached pronoun but it cannot head a gapped relative clause, and in the other case, an attached pronoun is not possible but relativization is. This suggests that the assumption made above may not be reliable, and further data to be examined in chapter 4 will show that this is indeed the case. The non-subject arguments of undergoer subject clauses present a more complex picture. They can be direct arguments if realized morphologically, and as obliques they can be licensed in two ways, by a preposition or by adjunction to the verbal head. Chapter 3 will present data for another type of clause which shows that this second possibility is more widespread in Indonesian, and can be the only possible means of realizing an oblique argument in one case. Undergoer subject forms of transitive verbs therefore use various strategies for licensing their non-subject arguments. Some of these are unusual cross-linguistically, and do not fit easily into the theory of argument realization set out in chapter 1. It is tempting to propose that there is some correlation between these facts and the fact that the subject argument of these verbs is directly specified by morphology, rather than being selected by the operations of LMT (except in the case of the bare verb actor subject clause type). This possibility will be pursued in chapter 5, when a fuller picture of Indonesian syntax is available. 103 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 3 ke- -an verbs This chapter deals with a group of predicates in Indonesian which are derived from other words by the addition of the circumfix ke- -an. This affix has several functions in Indonesian, but the syntactic behaviour of the words under consideration is distinctive and shows that they are verbs. The semantics of these verbs is also distinctive, almost always implying some bad effect on the subject: 1. Tomo kecurian mobil Tomo ke.steal.an car 'Tomo's car was stolen.' (IRG: 120) For this reason, these verbs have often been referred to as adversative verbs (e.g. Darjowidjojo 1978) or as adversative passives (e.g. Kana 1986) or accidental passives (Junus 1971)1. A clause such as example 1 has two nominals which meet the prima facie criterion for being direct arguments in Indonesian: they are not introduced by prepositions. The discussion of chapter 2 has already shown that this criterion is an oversimplification, and evidence presented below will show that the initial DP is a subject, but that the second DP does not have the properties expected of any type of direct argument in Indonesian. The first part of this chapter is a descriptive account of this type of verb, based largely on published sources. Of particular interest here is the issue of whether there are semantically-based sub-groups which can be identified within the group of verbs and, if so, whether these correlate with differences in syntactic behaviour. Section 3.2 examines previous accounts of the syntax of this clause type, particularly the analysis of Kana (1986, chapter 6) who treats them as clause unions in a Relational Grammar framework. Section 3.3 presents a new analysis, arguing that the predicate ke- -an exhibits highly unusual properties in the way in which a new argument structure is created from that of the base predicate with which it combines. This process is sensitive to the semantic relations between the base predicate and its arguments, and takes no account of the grammatical functions assigned by the base. The status of the bare DP second argument which some ke- -an verbs allow (see example 1) is also investigated in detail, and I conclude that it is an oblique argument. I claim, on the basis of this analysis, that the possibility of adjoining an oblique argument to the verb which licenses it is not restricted to the agent in undergoer subject clauses. The assumptions which I make in this work, described in chapters 1 and 2, lead me to reject these labels as inappropriate. These verbs are not the result of a lexical process which blocks the thematically most prominent argument from linking to a direct argument function and therefore they are not passive clauses. 1 104 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 3.1 A description of the ke- -an verbs 3.1.1 Functions of the circumfix ke-...-an The most common function of the ke- -an affix in Indonesian is to derive nouns, usually abstract. Thus baik 'good' has the nominalisation kebaikan 'goodness', and raja 'king' has the derived noun kerajaan 'kingdom'. This type of derivation is very productive. In section 4.1.2.8, such nominalizations of emotion and cognition predicates are discussed; of the 41 words for which data is presented, 31 have ke- -an nominalizations. A text example is the following: 2. Susila merasa dan berpendapatan, usahanya itu and hold.opinion effort.3 that Susila AS.feel semata-mata akan menimbulkan keamanan, solely FUT AS.float.CAUS NOM.safe.NOM kesentosaan, kemakmuran dan kecerdasan and NOM.shrewd.NOM NOM.quiet.NOM NOM.rich.NOM rakyat. populace 'Susila felt and believed that only that effort of his would bring about the safety, peace, prosperity and shrewdness of the people.' (Iskandar: 1946: 61, quoted in Teeuw 1977) This circumfix also derives two types of adjectives. There is a derivation of limited productivity, mainly restricted to colour words, with the meaning 'possessing to some degree the characteristic indicated by the base' (IRG: 50). The base is reduplicated in this derivation, thus biru 'blue' gives kebiru-biruan 'bluish' and ilmu 'knowledge' gives keilmu-ilmuan 'quasi-scientific'. This second example shows a tendency noted by Sneddon (IRG: 50) for these derivatives to have a negative connotation. The following is a textual example: 3. Meskipun jalan kendaraan itu amat cepat - maklum, although road vehicle that very fast know kegila-gilaan abang-abang supir tapi terasa older.brother.DUP driver ADJ.insane.DUP.ADJ but felt olehku lambat juga. leisurely only by.1SG 'Although the car's passage was very fast - you should know the chauffeur was a very crazy - but it felt slow to me.' (Iskandar 1952: 94, quoted in Teeuw 1977) The second type of adjectival derivation takes adjectival bases and intensifies them: kecil 'small' gives kekecilan 'too small'. This function has limited productivity in Indonesian. Sneddon ascribes this derivation to Javanese influence (IRG:50): the affix also exists in Javanese and is fully productive in this function in that language2. 2 Teeuw (1977) does not mention this function at all. This may be due to the fact that his data are all drawn from the work of an Indonesian writer of a previous generation, who 105 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 It is not easy to draw a boundary between the various functions of the circumfix ke- -an described above, or between at least the last of them and the verbal(izing) function which is the focus of this chapter. IRG (50) gives the following as an example illustrating the last function of the affix mentioned in the previous paragraph: 4. Kamu kebanyakan uang 2 ke.much.an money 'You have too much money.' The structure of this clause suggests strongly that it is not headed by an adjective. Adjectives in Indonesian do not generally take nominal arguments in this fashion (see discussion in section 4.1.3), and the clause seems to be semantically compatible with the adversative verb pattern: on this account the meaning would more accurately be 'You suffer from much money'. Other sources include this word as a verb (Junus 1971), and indeed in a later paper, Sneddon uses this example to demonstrate the spread of the pattern to include derivations from new bases (Sneddon 2000). None of the sources mentioned thus far gives explicit criteria by which the verbal words formed on this pattern can be distinguished from the words belonging to other categories. Teeuw (1977) quotes the following passage to show the extent to which categorisation is flexible even within a short passage: 5. India telah kehilangan seorang pemimpin besar. Dan India PERF ke.lost.an one.CLASS leader large and kehilangan bangsa India itupun terasa oleh ke.lost.an nation India that.EMPH felt by Indonesia sebagai kehilangan sendiri. Indonesia as ke.lost.an self 'India has suffered the loss of a great leader. And that loss of the Indian nation is felt as its own loss by Indonesia.' (Iskandar 1952: 425, quoted in Teeuw 1977) The structures in this passage show at least how to differentiate verbs from nouns. Each of the instances of the word kehilangan in this passage is followed by a DP, however the syntactic relations involved are different. In the first instance, kehilangan is the predicate of the clause: it is preceded by the auxiliary telah. In the second instance, kehilangan heads the subject DP, and the third kehilangan heads the complement of the preposition sebagai. In the second and third cases, the following DP could be replaced by the bound pronoun -nya functioning as a possessive, but this is not possible in the first case: 6. 7. *India telah kehilangannya Dan kehilangannya terasa .... was a native of Minangkabau. IRG (50) notes that the function in question is not considered standard by all educated speakers. 106 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 The problem of distinguishing adjectives and verbs remains. This is not a simple problem in Indonesian, and is discussed in detail in section 4.1.3 in relation to another group of words. Some of the crucial tests to be discussed there are morphological, and are not applicable in this case as the ke- -an circumfix does not co-occur with other morphology characteristic of adjectives. The syntactic tests discussed in section 4.1.3 also are not relevant in this case. Even the apparently clearly adjectival ke- -an forms do not occur in adverbial phrases with dengan: 8. *Dia berkata dengan kemanisan 3SG speak with ke.sweet.an (FOR: 'He spoke too sweetly.') These forms also cannot directly modify a noun; the relativizer yang is always necessary: 9. gue *(yang) kemanisan ke.sweet.an cake REL 'the oversweet cake' itu that This evidence suggests that the ke- -an forms traditionally regarded as adjectives (IRG, Teeuw 1977) are in fact intransitive verbs. This should be borne in mind by the reader, but is not an important fact in relation to the arguments presented here. What is important is that there are a number of forms with the affix ke- -an which can have two nominal arguments and that this is impossible for Indonesian adjectives except where they have the comparative morpheme se- prefixed to them: 10. orang setinggi pohon person COMP.high tree 'a person as tall as a tree' 11. *orang tinggi pohon The discussion of this chapter will take it that the ke- -an forms which can take two nominal arguments are unequivocal verbs. The discussion will refer to single argument ke- -an verbs from time to time, but always with the implied qualification that the categorial status of these forms is less secure. 3.1.2 Subgroups of ke- -an verbs Various attempts have been made to divide the ke- -an verbs into subgroups on the basis of their semantics, their syntactic behaviour, their derivational history or some combination of these factors. The differences in methodology are no doubt largely responsible for the differences in results: Junus (1971) identifies seven classes of ke- -an verbs, Dardjowidjojo (1978, afterwards SSAKI) identifies six and Sneddon (2000, afterwards ALKV) finds three, of which the one has four subclasses and another two3. These classifications are all based largely on semantic factors (see discussion below). IRG identifies six subclasses which correspond fairly closely to the groups in Sneddon (2000). The addition of the higher-level grouping is new in the later work. 3 107 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 Hiorth (1976) considers the syntactic environment to be more important and identifies three classes: ke- -an verbs which are followed by an agent, those which are followed by a DP which is not an agent, and intransitive verbs. This is similar but not identical to the classification adopted by Kana (1986). This author is primarily interested in the grammatical relations assigned by these verbs, and therefore only identifies three classes on the basis of whether there is a second argument, and whether it is a DP or a PP. Junus bases his classification on the transformations which are possible for each class of verbs he identifies4, and while in some cases this method reveals important facts, it is often obscure and the empirical base is lacking in detail. The following discussion is based mainly on the classifications of Dardjowidjojo and Sneddon which are presented in more detail and with better supporting evidence. I draw attention to parallels with Junus's work where relevant. The classification of SSAKI is based on the semantic relation between the verb and its associated nominals. That of ALKV is primarily based on the semantic relation between the verb and its subject. There is therefore some overlap in the two classifications, the most significant being the separation of the small group of verbs which are related to perception verbs: 12. Root dengar dapat lihat tahu Gloss hear discover see know ke- -an verb kedengaran kedapatan kelihatan ketahuan Gloss can be heard, audible be found out, detected can be seen, visible be found out, discovered Junus (1971) also identifies these as a group. The perceived entity is always the subject: 13. Rumahnya kelihatan dari sini house.3 ke.see.an from here 'His house can be seen from here.' (ALKV: ex32) ALKV notes that the perceiver is always implicitly present even when unexpressed, and that these verbs can take a subjectless clause as complement if the subject is human: 14. Dia kedapatan menggelapkan uang negara money state 3SG ke.discover.an AS.dark.CAUS 'He was caught embezzling state funds.' (ALKV: ex29) The perceiver is often introduced by the preposition oleh 'by', but this is only obligatory with pronouns and when the perceiver follows a complement clause. This class is not semantically uniform. kedapatan and ketahuan carry the implication that the event has adverse consequences for the subject if it is human, whereas kelihatan and kedengaran have no such implication. They have instead an abilitative meaning (ALKV: 524), and can be followed by an adjectival complement. Junus uses the term transformation, in the sense of Harris (1965) rather than that of Chomsky (1957 and elsewhere). 4 108 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 The other area of significant overlap between the two classifications is in the treatment of verbs with a single argument. ALKV treats all of these in a large group which includes all the verbs having a subject with the thematic role of patient and which suffers adverse consequences. This group also includes a set of verbs with two arguments, and the single argument verbs are subclassified according to the category of their root. SSAVI recognises two classes of one argument verbs and does not group them together, nor does it link either or both classes with any two argument verbs except in the overall grouping. The second group which is discussed in SSAVI may have only a single member, kehujanan 'caught in the rain' derived from the root hujan 'rain', which can be either a noun or a verb. ALKV treats this as a one argument verb based on a noun, while SSAVI takes it to be based on an argumentless ambient verb. The remaining verbs, that is the two argument verbs which are not perception verbs, are where the classifications of ALKV and SSAVI diverge. ALKV identifies two groups. Firstly, there are two argument verbs with a subject which suffers adverse consequences from the action referred to by the verb such as keracunan 'be poisoned by/with' (group 1d). In this case, the second argument can be either the agent of the action: 15. Mas Begong kemasukan setan brother Begong ke.enter.an devil 'Mr Begong got possessed by a devil.' (SSAVI: ex6) or it can be the medium through which the action was carried out: 16. Saya keracunan air basi 1SG ke.poison.an water spoiled 'I was poisoned by bad water.' In the classification of ALKV, this group is on the same level as all the oneargument verbs. Together they form the first-level class of verbs with patient subjects, and some verbs occur in both one-argument and two-argument groups. The second class of two-argument verbs is a first-level group. The subject is not a patient but rather is the possessor of the other nominal. The subject suffers adverse consequences as a result of something happening to the second argument: 17. Ali kematian ayahnya Ali ke.die.an father.3 'Ali suffered the death of his father.' (ALKV: ex20) Clauses with verbs of this type can be (partially) paraphrased by a clause containing the root verb and the two nominals in a single DP as possessor and possessed: 18. Ayah Ali mati father Ali die 'Ali's father died.' (ALKV: ex21) 109 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 This relationship was also noted by Junus (1971). ALKV divides this class into two subclasses on the basis of the thematic role of the second argument. In one subclass, this argument is affected by the action as in example 13, and is termed a patient in ALKV (group 2a). In the other subclass, the second argument is transferred by the action and is therefore termed a theme by ALKV (group 2b). All the verbs of this second class denote stealing: 19. Saya kecurian sepeda 1SG ke.steal.an bicycle 'I had my bike stolen.' (ALKV: ex22) The partial paraphrase for this subclass uses an undergoer subject verb in the 'agentless passive' clause type. Again the possession relation is made explicit in the paraphrase: 20. Sepeda saya dicuri bicycle 1SG US.steal 'My bike was stolen.' (ALKV: ex23) This group is not identified at all by Junus (1971). SSAVI also identifies three groups of two-argument verbs, other than the perception verbs, but they do not correspond exactly with the ALKV groupings. For SSAVI, the subject of all three groups is a patient. In one case, the second argument is an agent; this group includes words such as kemasukan (see example 15 above) and corresponds to part of ALKV's group 1d. For the other two groups identified by SSAVI, both arguments are claimed to be patients5. The difference between the two groups lies in the nature of the verb. In the first case, the verb denotes a process and in the second it denotes an action. The second group corresponds closely to ALKV's group 2b, but the examples given in SSAVI suggest that its first group corresponds to part of ALKV's group 1d. Nothing in SSAVI's classification corresponds to ALKV's group 2a, but ALKV's group 1d is not a unitary group in the classification of SSAVI. SSAVI makes no claim for any relationships between the six groups it identifies, whereas ALKV's classification is hierarchical. Figure 3.1 sets out these relations in a graphical format. SSAVI's terminology is based on that of Chafe (1970) and the extension of term patient may not be identical to that of the same term in ALKV. 5 110 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 ALKV 1a kelaparan 'suffer hunger' kelaparan kehujanan salju SSAVI 1 51 b kehujanan 'be rained on' c ketiduran 'oversleep' d kejatuhan 'be fallen on' 2 kemasukan 'be possessed by' 6 2a kematian 'suffer the death of' b kecurian 'have stolen' 3 3 kedengaran 'be audible' 4 Table 3.1 - Classifications of ke- -an verbs 1. SSAVI treats kehujanan salju 'to be snowed on' as a compound verb. ALKV argues against this position. 111 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 The analysis to be developed below will treat ke- -an verbs as complex predicates, and will argue that the semantic relations between the base predicate and its arguments are crucial to determining the argument structure of the ke- -an verb. Therefore it is important to retain all the distinctions which can be observed in the data, and SSAVI's observation that some two argument ke- -an verbs take an agent as the second argument will be significant. On the other hand, the category of the base of one argument verbs is not important; if the base assigns a thematic role it must be compatible with that assigned by the other part of the complex predicate and if the base assigns no role, only that assigned by the other predicate will be present. I therefore assume the following set of possibilities6: 21. Number of Arguments 1 argument 2 argument Argument 1 Patient Patient Patient Possessor Possessor Theme Argument 2 Theme Agent Patient Theme Perceiver One fact is made clear in this tabulation: although almost all ke- -an verbs imply that the subject is affected by the event, only two groups of the ke- -an verbs have thematic profiles which one would normally consider typical of passive verbs: the third and the last of the possibilities listed in 21. Therefore, in addition to the syntactic reasons for rejecting passive as an appropriate label for these verbs, it is also not particularly appropriate on semantic grounds. 3.2 Analyses of clauses with ke- -an verbs Only two of the sources discussed in the previous section go beyond offering a taxonomy of ke- -an verbs, and in each case the analysis that is offered is based on the idea that these verbs have a dual nature. The framework employed by SSAVI treats the relation between the verb and each of its arguments as semantically distinct. The verb need not have identical semantic characteristics in its relation to each argument. Therefore, for each class except the single argument verbs, SSAVI has two semantically distinct verbs in the representation of each clause. One of these is the root and the other is the root plus the circumfix. For example, SSAVI analyses the verb kejatuhan 'to be fallen on' as consisting of the verb jatuh which is a process and has a patient which is affected by the process (the object which falls), and the verb kejatuhan which is a state and has also a patient which is the entity adversely affected by the event. This analysis makes no claim about the syntactic relations between the verb(s) and the arguments; only semantic relations are considered. Patient here is an entity directly affected by the event and Theme is an entity which participates in the event without causing it or being affected by it. 6 112 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 3.2.1 The analysis of Kana (1986) Kana (1986, chapter 6) takes essentially the same idea but presents it formally within the framework of Relational Grammar. Clauses with ke- -an verbs have a single verb in their surface string, but have two predicates in some level of representation. In RG terms, they are examples of clause union. The circumfix ke- -an is a predicate in its own right in this analysis, having a meaning something like 'befall, affect adversely' (Kana 1986: 185). The root of the verb and its arguments constitutes the inner predicate sector, and ke- -an is the outer predicate, initializing relations in a later stratum. The relational diagram for a typical clause is as follows7: 22. 23. Ali kematian anak Ali AFFECT-die child 'Ali had a child die on him.' (Kana 1986: 196, ex2' [original gloss]) The predicate mati 'die' is initialized in the innermost stratum. It is an unaccusative verb (according to Kana - see discussion below and in section 4.3) and therefore it also initializes a 2 arc. In the next stratum, the union predicate ke- -an is initialized, forcing the base predicate mati into chômage. The union predicate also initializes a 2 arc and the Stratal Uniqueness Law (Perlmutter and Postal 1983) does not allow two arcs to bear the same relation in a single stratum. Therefore the 2 initialized by the base predicate is also forced into chômage. The Final 1 Law (Perlmutter and Postal 1983) requires every clause to have a 1 arc (a subject) in its final stratum, and this forces the 2 initialized by the union predicate to be revalued in the final stratum. Kana exhaustively treats the various arrays of relations which are possible in the initial stratum given her assumptions. There are seven possibilities, two of which are unattested in Indonesian according to Kana. One of these, that a single nominal heads two arcs in the initial stratum (multiattachment in RG terminology) does not occur in the language at all according to Kana and this possibility is ignored. Another possible initial stratum, with an unergative verb apparently does not exist in Indonesian either. However, this claim rests on Kana's analysis of intransitive verbs (1986 chapter 3), which is problematic. I return to this issue below in section 3.2.2.2 This discussion follows that of Kana 1986 in assuming that clause unions are monoclausal, multi-predicate structures (Davies and Rosen 1988), rather than multiclausal structures. As far as I am aware, nothing in the discussion hinges on this choice. 7 113 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 Base Predicate Type Predicate only Unaccusative Transitive2 Transitive Initial Relations Ø 2 2 (1) 2 12 Relation union 21 to Final relation Ø 1 2-Chômeur 2-Chômeur 1-Chômeur 1 2U = ≠ 2U ≠ 2U 2 = 2U Example kehujanan ketiduran kematian kecurian kedengaran Table 3.2 - Types of ke- -an clause according to Kana (1986) 1. 2U = union nominal, initial 2 of union stratum 2. (1) = unspecified 1 argument, semantically present but not expressed 114 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 (and see also section 4.3). The argument initialized by the union predicate ke-an will always be the final 1, therefore the interest of Kana's analysis is in the fate of any arguments initialized in the inner stratum. The possibilities are summarised in Table 3.2 (I subdivide one of Kana's categories for clarity). Various aspects of Kana's analysis are made obvious in this presentation. Firstly, an argument of the base predicate can only be the final 1 when it is identified with the 2 initialized by the union predicate. Any other argument of the base predicate will be a chômeur. Secondly, it is claimed that two types of chômeur occur in the final strata of various clause types. The evidence for this claim is examined in section 3.2.2.1. There are three areas in which Kana's analysis is problematic. Firstly, the claim that there are two possible second arguments which occur with ke- -an verbs is dubious. Secondly, the claim that unergative verbs cannot be the predicate of the innermost stratum is unmotivated and apparently contradicted by the data. And finally, Kana makes no attempt to capture the fact that in many cases the subject of the ke- -an verb is in a possession relation with the second argument. I now deal with each of these problems in turn. 3.2.2 Problems with Kana's analysis 3.2.2.1 THE SECOND ARGUMENT Kana divides the properties of the second argument of two-argument ke- -an clauses into two groups (1986: 200-201). The first group includes two properties, that the position of the argument is postverbal, and that it has no preposition. These are properties of 2s in transitive clauses for Kana. The second group of properties includes the fact that the second argument cannot be a bound pronoun: 24. *Ali kematiannya Ali ke.die.an.3 (FOR: Ali had him die.' (Kana 1986: 200, ex30) and the various properties associated with the fact that these verbs do not participate in the transitive system described in chapter 2. Thus, they cannot occur with the prefix meN-: 25. *Ali mengematian anaknya Ali child.3 AS.ke.die.an (FOR: Ali had his son die on him.') (Kana 1986: 200, ex31b) or any of the other clause types discussed in chapter 2. On this basis, Kana suggests that these arguments should be taken to be 2-chômeurs: they have some of the properties expected of final 2s, but they also lack some crucial properties. There are however some ke- -an verbs which take the preposition oleh before their second argument (actually only three verbs are listed by Kana for this group: kelihatan, kedengaran, and kedapatan). This preposition is one means of expressing the actor in an undergoer subject clause. Such actors are 1-chômeurs in Kana's analysis (1986, chapter 4) and the second arguments of these ke- -an verbs must be the same. 115 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 For reasons discussed in chapter 1, I do not consider some of Kana's evidence on this point relevant, and on some other points her data is incomplete. The discussion of section 2.2.2.1 has shown that distribution of bound pronouns is conditioned by complex factors and it is not clear yet whether ke- -an verbs should be expected to allow them. This point will be clarified in chapter 5, when the distribution of bound pronouns is established. Postverbal arguments not introduced by a preposition may be direct arguments in Indonesian, but are not necessarily so (see argumentation in section 2.5.4). The most important syntactic tests are whether or not the DP can float a quantifier, and whether it can be extracted to head a relative clause. The second arguments of all two-argument ke- -an verbs fail these tests. For each type of two-argument verb, the second argument cannot float a quantifier: 26. *Saya kecurian mobil-mobil itu kemarin 1SG ke.steal.an car.DUP that yesterday (FOR: I had all those cars stolen yesterday.) semua all 27. *Saya kematian anak-anak kemarin semua 1SG ke.die.an child.DUP yesterday all (FOR: I suffered the death of all my children yesterday.' 28. *Mobil itu kelihatan polisi kemarin semua car that ke.see.an police yesterday all (FOR: The car was visible to all the police yesterday.) And for each type, the second argument cannot head a relative clause: 29. *mobil yang saya kecurian car 1SG ke.steal.an REL (FOR: 'the car that I had stolen') 30. *anak yang saya kematian 1SG ke.die.an child REL (FOR: 'the child that I had die') 31. *polisi yang mobil kelihatan car ke.see.an police REL (FOR: 'the police to whom the car was visible') The second argument also cannot be fronted in the constructions discussed in section 2.2.2.2: 32. *Mobil itu, saya kecurian car that 1SG ke.steal.an (FOR: 'That car, I had stolen (from me).') 33. *Mobil itu saya yang kecurian car that 1SG ke.steal.an REL (FOR: That car, it was me who had it stolen.) 116 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 This evidence suggests that not only is the second argument of these verbs not a direct argument, but also that it is not available to any syntactic processes. Kana's description of the verbs which according to her have a 1chômeur as their second argument is also misleading. Her statement is the following: For examples such as 5 [an example with kedengaran. SM], the postverbal nominal is preceded by oleh, the regular marker of 1-chômeurs in passives. But the preposition is not obligatory in all cases. It is obligatory if the second argument is a pronoun, or if the second argument does not immediately follow the verb (ALKV: 524-525): 34. 35. Dia ketahuan (oleh) polisi mencuri barang itu 3SG ke.know.an (by) police AS.steal goods that OR: Dia ketahuan mencuri barang itu oleh polisi 3SG ke.know.an AS.steal goods that by police 'He was caught stealing those goods by the police.' (ALKV: ex33 Mobil itu kedengaran *(oleh) car that ke.hear.an by 'The car could be heard by me.' saya 1SG These conditions have a great deal in common with those that restrict the appearance of actor DPs in undergoer subject clauses - see section 2.5 for details. It is thus by no means as clear as Kana claims that 1-chômeurs and 2chômeurs display different syntactic behaviour. It is also the case that oleh may be used with the second argument of some other ke- -an verbs. ALKV has the following example: 36. Apalagi mereka pun keduluan oleh Speed ke.precede.an by Speed furthermore 3PL EMPH 'Furthermore, they were beaten to it by Speed.' (ALKV: ex36) and Verhaar (1984b) gives examples with keracunan 'be poisoned', kebanjiran 'be flooded' and ketinggalan 'be left behind'8. One additional point can be made about the behaviour of the second arguments of ke- -an verbs. It is only the perception verb group which allow the second argument to be moved away from the verb at all. For the other two-argument types, the intervention of any material (except for oleh in the cases discussed above) between the verb and the second argument is impossible: Verhaar (1984b) claims that in the case of ketinggalan the preposition is possible because the second argument is an experiencer. For him, this verb then groups with kelihatan and the perception verbs. The suggestion seems semantically unconvincing. 8 117 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 37. *Saya kecurian dulu sepeda 1SG ke.steal.an before bicycle (FOR: 'I had my bike stolen previously.') 38. *Dia kedatangan dulu ibunya 3SG ke.come.an before mother.3 (FOR: 'He was visited unexpectedly by his mother previously.') Again, this is reminiscent of the behaviour of actor DPs in undergoer subject clauses. To conclude the discussion of this point, the evidence shows that the second argument of the ke- -an perception verbs has different properties to those of the other verbs, but these properties are not so different as to allow us to conclude on syntactic grounds that the arguments have a different status. 3.2.2.2 RESTRICTIONS ON BASE VERBS Kana claims that only unaccusative intransitive verbs are possible as the base predicate with which ke- -an combines (1986: 208-211), although she can find no motivation for this restriction. There are however apparent counter-examples to this claim. SSAVI classifies the verbs kedatangan 'be visited´and kemasukan 'be entered, be possessed (by a devil)' as having an agent as second argument, and they are apparently derived from the intransitive verbs datang 'come, arrive' and masuk 'enter, go into' which intuitively have agentive subjects. Kana bases her classification of Indonesian intransitives on morphological criteria (1986, chapter 3, see also discussion in section 4.3) and notes that some Indonesian unaccusative verbs, including the two under discussion here, seem to have actor arguments. Relational Grammar after Rosen (1984) allows such mismatches, and Kana does not consider them as invalidating her hypothesis. The two most important tests that Kana uses for identifying unaccusatives are -kan suffixation (1986: 55-68) and peN- -an nominalizations (1986: 68-77)9. datang and masuk do display behaviour similar to that of Kana's other unaccusatives on these tests. Both use -kan as a causative suffix rather than an applicative: 39. Polisi mendatangkannya police AS.come.CAUS.3 'The police had him brought in.' (E&S: 132) 40. Ia memasukkan barang di lemari 3SG AS.enter.CAUS goods cupboard LOC 'She put the things in the cupboard.' (E&S: 364) And masuk has a derived nominal with the peN- - an affix: The capital N in peN- indicates an underspecified nasal segment which exhibits the same allomorphy as that in the actor subject prefix meN-. The two prefixes are undoubtedly linked historically. 9 118 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 41. pemasukan 'entering, registration, import, introduction' Against this, both of these verbs allowed agent-oriented adverbs to occur with them: 42. Dia datang ke Australia dengan 3SG come to Australia with 'He came to Australia deliberately.' sengaja deliberate I will argue below that this type of ke- -an verb is in fact derived from a transitive base predicate, with deletion of an applicative suffix in the morphological component. On this basis, Kana's claim that there are no unergative verbs which serve as base for ke- -an derivations is correct, but still unmotivated. A semantic reason will be suggested for the constraint in section 3.3.1. 3.2.2.3 POSSESSOR SUBJECTS Another weakness in Kana's treatment is also revealed by Table 3.2. The two cases in which the 2 argument of the base predicate is not identical to the argument initialized by the union predicate ke- -an coincide with the cases in which the final subject is the possessor of the entity affected by the base predicate. The generalisation is that there are two possible derivations for a base predicate which takes a patient or theme argument. Either the undergoer of the base is identical to the undergoer of the union predicate and it will be the final subject, or it is not identical to the union nominal, in which case it must be the possessor of that entity. Kana discusses the possibility of a possessor ascension analysis for the ke- -an verbs of this type, but rejects it because it cannot be used for all the ke- -an verbs (1986: 225-226). But in the cases where a possessor is the final subject, the relationship between the two entities is an entailment of the ke- -an verb. For example, with the verb kecurian 'to have stolen', it is entailed that the stolen object was in the possession of the final subject at the time of the stealing event. It is possible to utter the following sentence: 43. Saya kecurian uang itu, tapi bukan 1SG ke.steal.an money that but NEG 'I had the money stolen, but it was not my money.' uang money saya 1SG but it must still be the case that the money was stolen from the speaker; it was in their possession at the relevant time. Kana's analysis does not capture this fact about the semantic content of some ke- -an verbs. However, the generalisation just noted makes it possible to delimit precisely the cases in which possession is a relevant factor, and to characterise them semantically. In every case in which the base predicate takes a patient argument, the final subject of the ke- -an verb is affected by the event denoted by the base, either directly by being identical to the original patient, or indirectly by being the possessor of the original patient or theme. It is clearly desirable for an analysis of these verbs to capture this generalisation. 119 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 3.3.3.4 SUMMARY This section has argued that Kana's (1986) analysis of ke- -an verbs as clause unions, whilst based on a correct intuition, is deficient in three respects. Firstly, the evidence for the existence of two distinct grammatical relations among the second arguments of ke- -an verbs is not reliable. While there is a definite semantic difference between the second arguments of various verbs, they should nevertheless be given a unified syntactic treatment if possible. Secondly, the syntactic evidence for the status of certain verbs as unaccusative is not convincing enough to override the semantic intuition that a verb like datang 'come' has an agent argument. And lastly, Kana's analysis misses important semantic generalisations about the correspondence between the argument array of the base predicate and that of the derived verb. In the following section, I propose an analysis in the LFG framework which attempts to remedy all these defects. 3.3 A new analysis The analysis presented here adopts the insight that ke- -an verbs are the result of the composition of two predicates. The principle of lexical integrity is upheld by LFG, and therefore the composition of the base predicate and ke- an must be a lexical process in an LFG analysis, rather than being a syntactic operation as in the Relational Grammar analysis (see Alsina 1997 and Andrews & Manning 1999 for LFG treatments of complex predicates). Nevertheless, the essential insight is the same. The discussion of the previous section suggests a better sub-classification of the ke- -an verbs, based on the semantics of the two predicates and of their arguments, may reveal more accurate semantic generalisations which reveal how that composition operates. Section 3.3.1 presents arguments that the cases where the nature of the base predicate is not clear can be resolved by assuming that derived verbs with the applicative suffix -i can be the base for ke- -an derivation, but that the suffix is deleted in the process. Section 3.3.2 sets out in detail how the argument structure of ke-an verbs is derived from that of the base predicate on semantic principles, including a discussion of the complex case of the verb ketinggalan 'be left behind'. This section concludes with a proposal for the lexical entry of the predicate ke- -an. Section 3.3.3 examines the syntax of clauses with ke- -an verbs, paying particular attention to the question of what grammatical function is assigned to the second argument of these verbs. I agree with the explicit characterisation of the ke- -an morpheme given by SSAVI and Kana (1986) as having a meaning of 'undergo, affect' with a strong implication that the effects are adverse10. I also agree with Kana that the ke- -an morpheme constitutes a predicate in its own right, and that it takes an undergoer argument as its subject11. We would immediately predict from general principles that such a predicate ought to be intransitive: by virtue of its semantics, this predicate has an argument from the bottom of the thematic 10 The other sources cited accept some version of this position implicitly. The existence of the perception group requires that the statement be in this form. If these verbs are excepted, a stronger claim can be made: ke- -an takes an affected patient as its subject. 11 120 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 hierarchy, therefore any second argument associated with it would necessarily be more thematically prominent and would be expected to link to the subject grammatical function. The evidence discussed in section 3.3.3 will confirm this prediction, showing that the second argument must be analysed as an OBLθ in the LFG theory of grammatical functions. 3.3.1 What are the base predicates? In order to analyse the relation between the argument structure of the base predicates which derive ke- -an verbs and the final argument structure, it is first necessary to establish exactly what the base predicate is in at least some cases. Where the base takes no arguments, as in the case of nouns, there is no issue. Nor is there a problem in the case of one argument base predicates such as adjectives and a few verbs. However, what is not immediately clear is whether certain ke- -an verbs derive from a root word, or whether there is an intermediate derived predicate which feeds ke- -an affixation. It is always the root which is affixed with ke- -an, but I will argue here that there is evidence in at least one group of verbs that an intermediate stage is involved. The argument is clearest for the several ke- -an verbs which have alternative argument-taking properties. For example, the root racun 'poison' is a noun and derives a ke- -an verb which has only a subject. This is as would be predicted: a root which takes no arguments combines with another predicate which takes at least a subject, and the result is a oneargument verb: 44. Saya keracunan ke.poison.an 1SG 'I was poisoned.' But this verb can also appear in a clause with two arguments: 45. Dia keracunan permen basi spoiled 3SG ke.posion.an candy 'He was poisoned by spoiled sweets.' (ALKV: ex16) The semantics of the one argument clause are exactly parallel to those of other clauses with ke- -an verbs based on nouns. A literal rendering might be 'He was adversely affected by poison', just as kemalaman, from the noun malam 'night' means 'be adversely affected by night (be overtaken by night)' and kesetanan, from the noun setan 'devil', means 'be adversely affected by a devil (be possessed by an evil spirit)'. But the meaning of example 45 is rather different and does not correspond in any obvious way to a structure with a noun base. There is no compound noun structure meaning 'spoiled sweet poison', and even if there was, then it would be expected to be the input to the ke- -an affixation because, at least as a nominalising circumfix, ke- -an can attach outside compounds: 46. terus straight tidak neg terang direct jelas clear keterusterangan candour, straightforwardness ketidakjelasan obscurity 121 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 There is, however, a derived verb rancuni 'poison someone or something' which has an argument structure closely related to that of keracunan: 47. Permen basi meracuninya candy spoiled AS.poison.APPL.3 'Spoiled sweets poisoned him.' Similar alternations can be observed with the roots hujan 'rain' and banjir 'flood', the data for hujan being especially enlightening. The root can be either a noun or an ambient verb which takes no arguments. It also serves as the base for a ke- -an verb which can occur with one argument or with two: 48. Kami kehujanan 1PL.EXCL ke.rain.an 'We were caught in the rain.' 49. Dia kehujanan batu 3SG ke.rain.an stone 'He was showered with stones.' (ALKV: ex14) The two-argument structure in this case cannot be based on a compound noun. The compound hujan batu does exist, but it means 'hail'. Additionally, the two-argument structure is almost impossible with a literal meaning because as a base predicate hujan takes no subject (Indonesian does not have expletive pronouns): 50. Hari ini hujan day this rain 'It's raining today.' Therefore, there is no potential candidate for the second argument position except the rain itself, which would be a redundant way of expressing the same meaning as example 48: 51. *Kami kehujanan hujan rain 1PL.EXCL ke.rain.an (FOR: We were rained on by the rain.) But, as with the root racun, there is a derived verb which has an argument structure identical to that of kehujanan: 52. Batu-batu menghujaninya stone.DUP AS.rain.APPL.3 'Stones rained on him.' In fact, all the verbs which ALKV lists in its group 1d, verbs with two arguments of which the first is a patient, appear to be derived from verbs with the suffix -i, with one exception discussed in section 3.3.1.2. There is independent evidence that morphological material can be deleted in Indonesian when a sequence of affixation occurs. For example, the adjective tinggi 'high' can be the base for a causative verb tinggikan 'raise, heighten'. There is also a derived noun peninggian (peN-tinggi-an) and 122 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 although this appears to take the root adjective as its base, the meaning is 'heightening, raising' which is clearly based on the causative verb even though the suffix -kan is not retained. A parallel example is provided by the adjective root pancar 'scattered'. The causative verb is pancarkan 'broadcast' and the nominalisation is pemancaran (peN-pancar-an) 'action of broadcasting'. This morphological process is very common in Indonesian, particularly in nominalizations of verbs. IRG (28) lists several examples for the prefix peN-, which derives nouns indicating the performer of an action12: 53. Base bangkit Gloss Caus V Gloss Nominal Gloss rise bangkitkan generate pembangkit generator satu one persatukan unify pemersatu unifier tenang calm tenangkan make calm penenang sedative kendali rein kendalikan control menara pengendali control tower There are also cases where a derived nominal may be ambiguous between a derivation from an unsuffixed verb and a suffixed verb (IRG: 40): 54. Verb tembak tembakkan Gloss Nominal penembakan penembakan shoot (s.o.) fire (a bullet) Gloss shooting (of s.o.) firing (of a bullet) and in at least one case, either of two suffixed verbs could be the base, but neither suffix appears in the derived nominal: 55. Verb terangi terangkan Gloss illuminate explain Nominal penerangan penerangan Gloss lighting explanation (The root terang means 'clear, bright') 13. A similar point is made by ALKV in relation to two ke- -an verbs, kegelapan 'overtaken by darkness' and kepedasan 'suffer from food which has too much chili'. These two verbs are apparently derived from adjectives, gelap 'dark' and pedas 'hot'. However, the other ke- -an verbs which are derived from adjectives can be replaced by the root adjective without drastic change of meaning: 56. Dia kelaparan 3SG ke.hungry.an 'He suffers from hunger.' 57. Dia lapar 3SG hungry 'He is hungry.' Some causative formations include the prefix per-. persatukan is such a verb, see IRG: 98103 for details. 12 13 See also footnotes 6 and 24 of chapter 4 for discussion of other relevant examples. 123 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 This is not the case for gelap and pedas: 58. Saya kegelapan 1SG ke.dark.an 'I was overtaken by darkness.' 59. ?Saya gelap 1SG dark 'I am dark.' For this reason, ALKV suggests that the ke- -an verbs are derived from nominalizations of the adjectives, also formed with ke- -an, kegelapan 'darkness' and kepedasan 'heat' with deletion of the doubled affix. It is not therefore implausible to suggest that the applicative suffix -i can be deleted from a verb when that verb is the input to the formation of a ke- -an verb, and this hypothesis does account for the argument structure which is available to the predicate composition process. The parallel between some ke- -an verbs and undergoer subject verbs with the suffix -i has been noted by previous studies (ALKV, Cartier 1978, and SSAVI although in that case the relevance of the parallel is denied). The difference between the two possibilities is the adversative meaning which is present in the ke- -an clause, otherwise the two are propositionally similar: 60. Dia kedatangan wartawan 3SG ke.come.an journalist 'He was unexpectedly visited by a journalist.' (ALKV: ex11) 61. Dia didatangi wartawan 3SG US.come.APPL journalist 'He was visited by a journalist.' (ALKV: ex12) No doubt this is one reason for the term 'adversative passive' being applied to the ke- -an clause type. The arguments of this section have been directed to establishing that one group of ke- -an verbs are derived from transitive verbs with the applicative suffix -i, even though this suffix does not appear in the derived form. Establishing this has three consequences for the discussion of the following section. Firstly, the issue of whether there are single argument verbs with an agent which function as the base for ke- -an verbs is irrelevant: the dubious cases actually involve a base transitive predicate. Secondly, the ke- an verbs in question are the result of the composition of the predicate ke- -an and a predicate with two arguments. And thirdly, the second argument of the base predicate has a particular character. While generalisations about the functions of verbal morphology in Indonesian are very hard to maintain, it is possible to say that the applicative suffix -i is most often associated with the promotion of a location to the object function (for detailed examination of the question, see Tampubolon 1983 and Voskuil 1996). This is clearly true of the majority of the verbs in ALKV's group 1d. It follows from this, that the objects of transitive verbs with the -i suffix are not conceptualised as affected entities. It is certainly possible for them to be affected by the event, but this is 124 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 incidental. Their essential role is as the location of the event. A clear example of this can be seen with the verb jatuhi 'fall on'. It can be used with an object which, on the basis of real world knowledge, can be said to be affected: 62. Kelapa itu menjatuhi kepala coconut that AS.fall.APPL head 'The coconut fell on my head.' saya 1SG But another lexical item can be substituted in the object position, and the apparent entailment of being affected is not present: 63. Kelapa itu menjatuhi atap rumah coconut that AS.fall.APPL roof house 'The coconut fell on the roof of my house.' saya 1SG Given that the roof in question is of some robust material, we do not assume from such an utterance that the roof was affected by the event. This fact about verbs with the suffix -i will be important in the following discussion. 3.3.2 Properties of the ke- -an predicate 3.3.2.1 DERIVING THE ARGUMENT STRUCTURE The unusual properties of the predicate ke- -an are best seen by considering first the verbs for which a new participant is added to the event, the possessor of the affected entity. There are two groups of these verbs. One group has intransitive bases, the other transitive bases, but in each case the subject of the ke- -an verb is a person in some relation to the entity affected by the event such that the person is also affected. The relation can always be syntactically expressed by a possessor-possessed structure, although the actual relationship may not be of that nature14. The entity affected by the event denoted by the base predicate is inanimate for almost all these verbs. the exceptions are kematian 'suffer the death of' and keguguran 'have a miscarriage'15. The entity affected by the base predicate in these cases is no longer animate as a result of the event; therefore the same generalisation does in fact apply to these two verbs also. Therefore we can say that the subject of these ke- -an verbs is the entity most affected by the event, where a human who experiences adverse effects is treated as more affected than an inanimate patient or theme. I shall refer to the added argument in the following discussion as an extended patient. The ke- -an verbs which have an extended patient as subject are an example of external possession in the sense of Payne and Barshi (1999). These scholars define an external possession construction as one in which: 14 This qualification is intended to cover the case of kinship relations in particular. keguguran never has an expressed second argument. ALKV argues (p521) that the semantic relation between the two arguments of kematian is parallel to the relation between the subject of keguguran and an understood second argument janin 'fetus'. The discussion assumes this position. 15 125 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 a semantic possessor-possessum relation is expressed by coding the possessor as a core grammatical relation of the verb and in a constituent separate from that which contains the possessum. (3) External possession constructions with an adversative meaning are attested in at least one other Austronesian language, Tukang Besi (Donohue 1999). Shibatani (1994) argues that external possession should be seen as part of a wider spectrum of constructions, including ethical dative and adversative constructions, which have in common the feature that they treat a participant that is not a semantic argument of the verb as a syntactic argument. He suggests that the non-thematic argument is integrated into the construction on a pragmatic basis, and that some readings, such as inalienable possession, are more easily achieved than others. ke- -an verbs do not fit into such an account well. Although some ke- -an verbs introduce an argument which is not semantically related to the base predicate, the account presented here claims that such arguments are semantically selected by the ke- -an predicate itself: it is a one-argument predicate which takes a patient as its subject. Almost all ke-an verbs are adversative in meaning, and the connection between this type of construction and external possession pointed out by Shibatani is important. It is therefore unsurprising that the one type of verb should allow the other type of meaning in some cases, but the possibility is more syntacticized in Indonesian than in some of the other languages discussed by Shibatani (see for example the comparison of Spanish and Japanese, examples 28 and 29 and related discussion). A case where the nature of the possession relation between two possibilities is significant is discussed in section 3.3.2.3, although even in that case the proposal given here accounts adequately for the contrast. There is a clear contrast between the ke- -an verbs formed from transitive verbs which have affected patients and those which, as argued in section 3.3.1, are based on transitive verbs whose second argument is a location. Where the base predicate is an -i suffixed verb, there is no possibility of the affectedness being extended to a possessor. This can be understood by recalling the argument presented in section 3.3.1 to the effect that affectedness was not entailed for the object of such verbs, but it is entailed for the subject of ke- -an verbs (with the exception of the perception group). This semantic requirement of the ke- -an predicate is exhausted in converting the non-affected second argument of a transitive verb into an affected subject. On the other hand, where there is a fully affected argument already available, the semantic force of the ke- -an predicate can be used to add a new affected argument (the sufferer in the terminology of ALKV). The proposal here is that the argument structure of ke- -an verbs is established on semantic grounds; the most affected entity in the situation described will be selected as subject of the ke- -an verb. The crucial data which confirms that the argument expression of ke- -an is semantically driven comes from two verbs both meaning 'to be addicted', kecanduan and ketagihan. The base forms of both these verbs take the addicted person as subject: 126 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 64. Ia telah mencandu minum kopi 3SG past drink coffee AS.addicted 'She has become addicted to coffee.' (E&S: 103) 65. Saya tagih rokok 1SG addicted cigarette 'I am addicted to smoking.' (E&S: 540) The ke- -an verb has the same subject, the addicted person: 66. Kamu sudah kecanduan cara roman yang past ke.addicted.an way romantic REL 2SG dilihat di film film US.see LOC 'You are addicted to the romantic ways which you have seen in the movies. (E&S: 104) 67. Ia ketagihan madat 3SG ke.addicted.an opium 'He is addicted to opium.' (E&S: 541) This would be highly surprising if the ke- -an verb selected its subject on the basis of the grammatical functions assigned by the base predicate. In all other cases where the base predicate has two arguments, the subject of the ke- -an verb is either the undergoer or a human related to the undergoer. But in this case, the subject of the ke- -an verbs is the actor of the base predicate. This is, however, unsurprising if the subject of the ke- -an verb is selected on semantic grounds. For these two verbs, the actor of the base predicate is the entity which suffers adverse effects from the event (more accurately, the state) which the base predicate denotes. This semantic theory allows us to give a principled account of the complete data. Firstly, there are bases which have no argument of their own. In this case, the derived verb has a single argument which is affected by the referent of the base, for example malam 'night' and kemalaman 'overtaken by darkness'. Secondly, the base can have a single argument which is the theme of a state, typically an adjectival base. The derived verb has the same single argument with the added meaning that it is affected by the state, for example lapar 'hungry' and kelaparan 'be affected by hunger'. Thirdly, the base can be a verb with only an affected patient argument. In this case, ke- -an can add an argument, an extended patient. Fourthly, the base can be a transitive verb with an affected patient. ke- -an can add an extended patient in this case, but it only allows the expression of two arguments and the agent remains unexpressed. This fact can be interpreted in two ways. Either it is a requirement of the ke- -an predicate that if the base predicate has a patient, this must be expressed in the final argument structure, or the arguments of the base predicate are ordered with respect to affectedness, and ke- -an takes its arguments in order from the lowest on this scale. The only evidence that seems to bear on this question is that two verbs can alternate between expressing only a patient as subject, and having an extended patient as 127 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 subject and the original patient as second argument. These are kebakaran 'catch fire' and ketinggalan 'be left behind': 68. Rumah saya kebakaran house 1SG ke.burn.an 'My house caught fire.' 69. Amat kebakaran rumah Amat ke.burn.an house 'Amat's house was burned down.' 70. Buku saya ketinggalan di perpustakaan book 1SG ke.leave.an library LOC 'I left my book in the library.' [LIT: My book was unintentionally left in the library.] (ALKV: ex8) 71. Saya ketinggalan buku di rumah 1SG ke.leave.an book LOC house 'I left my book at home.' (ALKV: ex27) This data is inconclusive but probably tends to support the first interpretation, expression of a patient of the base predicate is part of both structures. The fifth possibility is that the base is transitive, but the undergoer argument is not coded as affected, that is, it is one of the -i suffixed verbs discussed above. In this case, the ke- -an changes the semantics of the base predicate by making its undergoer an affected patient, and the possibility of an extended patient is not available. Similar arguments apply to two of the perception verbs, ketahuan and kedapatan. Both of these can plausibly be derived from transitive verbs with the suffix -i. dapat has various meanings as a root, mainly modal, but also including 'be found', but this is an intransitive verb. dapati, on the other hand, is a transitive verb meaning 'find, discover'. Similarly, the root tahu means 'know, be cognizant of' and ketahui means 'find out, discover'16. Both ketahuan and kedapatan 'always indicate something negative or unfavourable for [the subject]' (AKLV: 524). The remaining two perception verbs, kedengaran 'be audible' and kelihatan 'be visible' are the only two ke- -an verbs for which the semantic account is less convincing. The root in each case forms an unsuffixed transitive verb: 72. Dia mendengar suara ibunya 3SG AS.hear voice mother.3 'She heard her mother's voice.' 73. Kami sudah melihat 1PL.EXCL PERF AS.see 'We've seen that movie.' film film itu that The added morpheme ke- in the derived verb is due to the fact that this root is treated as monosyllabic. 16 128 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 The perceived entity is thus treated as a patient rather than a location, as far as verbal morphology is concerned, and in this case we would predict that an extended patient should be possible with the ke- -an verbs. This is not the case: 74. *Saya kelihatan mobil 1SG ke.see.an car (FOR: My car was visible.) It is unlikely that these two verbs are actually based on -i suffixed forms, as E&S do not list a dengari at all. Therefore, we are forced to conclude that these two verbs behave like those with a non-affected undergoer even though they do not have the morphology that this behaviour is usually associated with. I assume that this oddity is not unrelated to the fact that the semantic effect of ke- -an affixation for these two verbs is different from all other cases. For these verbs, the derived verb has an abilitative rather than an adversative meaning. There is still a semantic distinction between a ke- -an verb clause and the closest paraphrase, but it is not the same distinction as in the cases previously discussed: 75. Musik itu saya dengar hear music that 1SG 'The music was heard by me.' 76. Musik itu kedengaran oleh saya music that ke.hear.an by 1SG 'I could hear the music.' (ALKV: ex34) I offer no account here of the factors which cause these differences. This semantic account makes it easy to understand why there are no ke-an verbs with unergative intransitives as base. By definition, such verbs do not have an affected patient in their thematic structure. But the predicate ke- an requires an affected entity as its subject, and such an argument can only be added when the base has no arguments, or when an extended patient is part of the situation. Unergative intransitives meet neither of these conditions. If such a verb were combined with the ke- -an predicate, there would be no possible subject which fulfilled the semantic restrictions imposed by the predicate and therefore the lexical composition process must fail. 3.3.2.2 WHAT GRAMMATICAL FUNCTIONS? 3.3.2.2.1 Subject All the sources mentioned thus far agree on one point, that the DP which normally precedes the verb and whose referent is the entity which suffers adversely as a result of the event, is a subject. The tests described in section 1.1 all point to this conclusion. Thus, this position can be occupied by the third person pronoun ia: 129 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 77. ia telah kecurian barang-barangnya 3SG PERF ke.steal.an article.DUP.3 'his luggage was stolen from him' (Hiorth 1976: 78) The DP in this position can be extracted and can head a relative clause: 78. bayi yang ketularan penyakit dari para pedagang ke.infect.an disease from PL trader baby REL 'a baby that caught a disease from salesmen' (Hiorth 1976: 77) It is also be the case that this position can be controlled in a complement clause. Some examples are potentially ambiguous: 79. Aku mimpi kedatangan seorang asing dream ke.come.an someone alien 1SG 'I dreamt that an unknown person visited me.' (Hiorth 1976: 76) The gloss provided by Hiorth suggests that this is an example with a complement clause. But it is also marginally possible to construe kedatangan here as a noun, meaning 'arrival'. The gloss for the sentence would then be 'I dreamt of the arrival of an unknown person'17 This particular case is further complicated by the fact that the data discussed in section 4.1 include no example of the verb mimpi taking a complement clause; on the other hand they also do not include unequivocal examples of it taking a bare DP complement. Elicited examples can resolve the question easily. The word kehujanan has no nominal reading; it only means 'be rained on', and it can occur in a complement clause without an independent subject. Note also that kehujanan is negated by tidak in the following example, rather than by the nominal negator bukan: 80. Saya senang tidak kehujanan. 1SG happy ke.rain.an NEG 'I am happy not to be rained on.' The adversely affected argument of a ke- -an verb can control the reference of the subject of a predicative complement, and this provides additional evidence that such arguments are indeed the subject of their clause. Data which reinforces this argument is presented in section 3.3.2.2.2. 3.3.2.2.2 The status of the second argument - predicative complements It was argued in section 3.2.2.1 above that the evidence for treating the second arguments of ke- -an verbs as being of more than one type was less than compelling. I will therefore assume that all such arguments have the Native speaker judgment is that this meaning would be expressed by the following sentence: 17 Aku memimpikan kedatangan orang asing 1SG AS.dream.APPL NOM.come.NOM person alien 'I dreamt of the arrival of a stranger.' 130 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 same syntactic status, and turn to the question of what exactly that status is. LFG recognises only four grammatical functions that a verb can assign to a nominal argument: SUBJ, OBJ, OBJθ and OBLθ. The data presented in section 3.3.3.1 show that the preverbal argument is the SUBJ of a clause with a ke- -an verb. General principles of unification-based grammar prevent an attribute, such as a grammatical function, being assigned more than once in the relevant structure, therefore a clause can only have one SUBJ and the second argument of the ke- -an verb is not it. OBJ and OBJθ are both terms and it is a property of terms in Indonesian that they can launch floating quantifiers. As discussed in section 3.2.2.1, the second arguments of ke- -an verbs do not have this property, or any other property associated with term arguments except for appearing without a preposition. I therefore conclude that these arguments are not either OBJ or OBJθs. The only possibility which remains is that the second argument of a ke- -an verb is an OBLθ. This conclusion is supported by evidence from the ke- -an verbs which take a predicative complement as an argument. A number of verbs which serve as the base for ke- -an derivations can appear with a complement clause predicated of their subject (an XCOMP). This property is inherited by the derived verb, but in some cases there is a significant change to the control relation which establishes the reference of the missing subject. Some of the base verbs which allow predicative complements have the subject position of the complement controlled by the matrix object, and some have it controlled by the matrix subject. But in all cases, the subject of the derived ke- -an verb controls the reference of the subject of the complement. ke- -an verbs are formed by a lexical process, and this process affects the control relation, therefore this fact demonstrates that these are examples of lexically induced functional control (Bresnan 1982). It is a consequence of the LFG theory of control that 'the controllers of lexicallyinduced functional control relations must change under the lexical operations on functional assignments' (Bresnan 1982: 401). This clause type therefore provides additional evidence that the second argument of ke- -an verbs is a non-term, that is an OBLθ. There are base verbs which can take a SUBJ and an XCOMP as arguments, and in this case there is no change in the control properties as a result of the composition with the ke- -an predicate. For example, the verbs tagih 'addicted to' and candu 'addicted to' are of this type. By a lexical redundancy rule (Bresnan 1982), the following control equation is added to the lexical entries of these verbs: (↑SUBJ) = (↑XCOMP SUBJ). This allows clauses such as the following: 81. Ali tagih membaca koran Ali addicted AS.read newspaper 'Ali is obsessed with reading the paper.' (E&S: 540) When this verb is combined with the ke- -an verb, the subject of the base is the affected entity in the situation and therefore is also the subject of the derived verb. Therefore the control equation is unchanged: 131 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 82. ke- -an V 'undergo < x : P < ..... > , (y), (XCOMP) >' tagih V 'addicted < x : >' XCOMP (↑SUBJ) = (↑XCOMP SUBJ) ketagihan V 'suffer addiction < x : XCOMP > ' (↑SUBJ) = (↑XCOMP SUBJ) The resulting verb appears in a clause such as the following: 83. Dia ketagihan berjudi 3SG ke.addicted.an gamble 'He is addicted to gambling.' (ALKV: ex40) For both the base verb and the derived verb, the f-structure of the clause is of the following type: 84. The subjects of the two clause nuclei are identical. The other ke- -an verbs which allow a complement clause are derived from verbs for which the XCOMP subject is controlled by the matrix object. Where the XCOMP appears in the lexical entries of these verbs, it carries the annotation (↑OBJ) = (↑XCOMP SUBJ). This allows a clause such as the following: 85. Saya mendapatinya minum 1SG drink AS.find.3 'I found him drinking coffee.' with the associated f-structure: 132 kopi coffee NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 86. But where this lexical entry has served as the base for a ke- -an verb, the control properties of the derived verb are different. In this case, the subject of the derived verb controls the reference of the subject of the complement: 87. Dia kedapatan menggelapkan uang 3SG ke.find.an money AS.dark.CAUS 'He was caught embezzling state funds.' negara state The ke- -an verb must have the control equation (↑SUBJ) = (↑XCOMP SUBJ) with an f-structure of the type illustrated in 84. As noted above, it is a consequence of the LFG theory of control that 'the controllers of lexicallyinduced functional control relations must change under the lexical operations on functional assignments' (Bresnan 1982: 401). From this, it can be predicted that lexically-induced functional control relations will change under operations such as passive. This is indeed the case for a verb such as dapati: when the undergoer-subject form didapati is used, control is by the SUBJ, the OBJ of the actor –subject form. the following example only has a reading in which Ali is the coffee-drinker18: 88. Ali didapati minum kopi 1SG drink coffee AS.find.3 ‘Ali was found by Siti drinking coffee’ oleh by Siti Siti The other type of functional control relation allowed by the theory is constructionally-induced functional control. As the terminology implies, the relevant control equation is introduced as a functional annotation of a cstructure node in this case. Any such annotation cannot be affected by lexical processes. Therefore we can conclude that a verb such as dapati can lexically induce functional control. Additional properties of functional control can be derived from the LFG theory of control, two of which are relevant to this discussion. Firstly, The account of Pro-V clauses and di-V-nya clauses developed in Chapter 2 suggests that the control properties of a verb such as dapati might be ambiguous in such clause types. This is a question for further research; related Balinese data are discussed by Arka & Simpson (1998). 18 133 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 constraints on the lexical encoding of semantically restricted functions ensure that they can never be functional controllers (Bresnan 1982)19. And secondly, there is an unmarked hierarchy for choice of controller. If the lexical entry of a verb includes the function XCOMP, the controller will be the OBJ if there is one and it will be the SUBJ otherwise. In some cases, it is possible for a ke- -an verb to have two nominal arguments and an XCOMP also. If it were the case that the second nominal argument was a term, we would predict that it would control the reference of the subject of the complement clause. But this is not the case; the subject of the ke- -an verb controls the complement subject in these cases also: 89. Dia ketahuan polisi mencuri barang itu goods that 3SG ke.know.an police AS.steal 'He was caught stealing those goods by the police.' (AKLV: ex33) This is therefore additional strong evidence that the second argument of a ke-an verb is an oblique (and also that the preverbal argument is a subject). An alternative explanation might be that the ke- -an predicate itself has a lexical specification that its subject will be the controller for any open function which it might take as an argument. But this account would miss the generalisation that an XCOMP is only possible with a ke- -an verb if it is possible with the base verb: taking such a complement is a property of the base verb which is inherited in the lexical process which forms ke- -an verbs. 3.3.2.3 A LEXICAL ENTRY FOR KE- -AN The predicate ke- -an takes as arguments a subject, another predicate with its complete argument structure, and optionally an oblique argument and a clausal complement with a controlled subject position: 90. ke- -an V 'undergo < x : P < ..... > , (y), (XCOMP) >' Following Alsina (1997), I assume that the lexical process of predicate composition involves the identification of arguments of the embedded predicate with arguments of the inclusive predicate. For ke- -an, the possibilities for identifying arguments are constrained by the following principles: Bresnan (1982) treats OBJ2 as a semantically unrestricted function, while Bresnan and Kanerva (1989) and later work treat OBJθ as a semantically restricted function. This has consequences for the theory of control as set out in Bresnan (1982), but they are not relevant to the current discussion. Bresnan (1982) gives OBJ2 as the first choice for controller if that function is assigned by a verb; I omit this possibility from the discussion. 19 134 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 91. 1) The argument obligatorily associated with ke- -an is a patient and the most affected entity in the situation must be identified with it. 2) A more animate entity in a close relation (typically possessor) to an affected patient is an extended patient, and is the most affected entity in a situation. 3) An affected patient or theme must be identified with an argument of the inclusive predicate. I assume also that some general constraint against discarding information ensures that all the arguments of the base verb that can be realised will in fact be realised. The control properties of the XCOMP argument fall out from general principles; these were discussed in section 3.3.2.2.2. The working of the principles set out in 91 can be seen from the following examples. The noun malam 'night' has no arguments, therefore the lexical entry for the derived verb kemalaman will be the following: 92. kemalaman V 'be affected by night < x >' The subject function for ke- -an is obligatory, therefore a new argument is added, and the meaning of that predicate guarantees that this argument will be a patient. The (derived) verb jatuhi has the following lexical entry (subscripted numbers are added to assist keeping track of the identity of arguments): 93. jatuhi Vtr 'fall on < x1, y2 >' The second argument is a location not an affected patient, therefore principles 91.2 and 91.3 do not apply. The location is nevertheless the most affected entity and it therefore becomes subject. The oblique slot of the ke- -an verb is available to realise the first argument of the base verb, and the result is: 94. kejatuhan V 'be fallen on' < x2 : y1 > The verb mati 'die' takes a single argument which is an affected patient. Therefore principle 91.2 is applicable: a more animate entity, in this case a live person, in a possessor-like relation to that patient is an extended patient in this situation and the most affected entity. By principle 91.1, this person will be the subject and the oblique slot will be available to realise the argument of the base predicate: 95. 96. mati Vitr < x1 (possessed by y2) > kematian V < x2 : y1 > The other classes of verbs described previously can be derived in similar fashion. The a-structure of ke- -an verbs provides evidence that deriving a verb's valence from the interaction of its a-structure and the principles of LMT does not work in all cases. If the lexical entry of ke- -an does not represent the fact that it is an intransitive verb with a single direct argument, the version of astructure presented in Bresnan (2001a) and LMT make two incorrect 135 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 predictions, the first of which also depends on the analysis presented in chapter 2. Firstly, where the base predicate has an agent and a patient as arguments, the a-structure of the complete predicate would be the following, with linking to semantic roles included for clarity: 97. ke-V-an (agent) < x, (patient) y > | SUBJ This configuration meets the conditions for morphological adjunction of a pronoun as a direct argument: the verb has lexically specified linking of its second argument to the SUBJ GF (chapter 2, example 153). But this is not possible, as seen in example 24. The a-structure therefore cannot be as shown in 97. Secondly, in the case where an extended patient and the patient of the base predicate are the arguments of the ke- -an verb, the a-structure would be: 98. ke-V-an (extended patient) < x, | (patient) y > SUBJ Indonesian has second objects, and I have assumed that these are treated in the same way as in English. Therefore, we would expect the second patientlike argument in this configuration to be assigned the feature [+o], and be mapped to the GF OBJθ, giving a verb with the GFs SUBJ and OBJθ. But the evidence presented in section 3.2.2.1 has made it clear that the second argument of such verbs is not a term, therefore not an OBJθ. Neither of these cases is problematic if the information that ke- -an has a single term argument is represented in the a-structure. Only one term position is available, and principle 91.1 specifies which argument will be mapped to it. Any other argument can only be realised as an oblique. I take this to be important evidence in support of the version of a-structure advocated by Manning (1996a). 3.3.2.4 THE CASE OF KETINGGALAN ALKV groups the verb ketinggalan 'be left behind' with the verbs discussed in the section 3.3.1.1, those that have two arguments of which the first is a patient. But the range of clause types in which this verb can appear shows that this classification is unsatisfactory. Verhaar (1984b) treats ketinggalan as belonging to the same group as the perception verbs. This is semantically implausible, but ketinggalan can occur with a complement clause with controlled subject, which is a point of similarity. In fact, close examination of the data shows that this verb can appear in almost every clausal structure attested for ke- -an verbs. The account developed above forces the conclusion that this can only occur if there is more than one predicate which is serving as the base for ke- -an derivation, and this section will argue that this is indeed the correct solution. 136 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 The most immediate and striking contrast is between the following two possibilities. Firstly, ketinggalan can occur with an extended patient as subject, but also allows the true patient to occur in that position: 99. Saya ketinggalan buku di rumah 1SG ke.leave.an book LOC house 'I left my book at home.' [LIT: I suffered the leaving of my book at home.] (ALKV: ex27) 100. Buku saya ketinggalan di perpustakaan book 1SG ke.leave.an library LOC 'I left my book in the library.' [LIT: My book was unintentionally left in the library.] (ALKV ex8) The same verb can be used in a structure parallel to example 99, but without the possibility of the alternative: 101. Saya ketinggalan bis ke.leave.an bus 1SG 'I missed the bus.' (LIT: I was left behind by the bus.) (ALKV: ex17) 102. *Bis saya ketinggalan This contrast might be argued to be due to the factors discussed by Shibatani (1994). The relation between a book and its possessor, and a bus and one of its passengers, even when we can say my bus, is sufficiently different that one case might allow the possessor to be integrated into the scene and the other might not. However, in this second case, the second argument is an actor: it is the bus that leaves and thereby affects the subject. Following the arguments made previously, an extended patient is only possible when the base predicate has an affected patient as one of its arguments, and therefore the base predicate for examples 99 and 100 must be of this type. It will have the entity left behind as patient. On the other hand, the base predicate in example 101 will not have an affected patient in its argument structure. An appeal to pragmatic factors is not necessary, even in this case. There are two possible bases for the ketinggalan seen in examples 99 and 100. The intransitive verb tinggal means 'stay, be left', it plausibly has a patient as its sole argument, and the meaning is compatible. On this account, the literal meaning of example 99 would be 'I suffered because the book stayed at home.' The alternative is that the causative verb tinggalkan is the base. This verb is normally translated as 'leave behind', the more literal meaning is 'cause to stay, cause to be left'. This is a transitive verb with an agent, the causer, as well as the affected patient inherited from the root verb, and if this were the base predicate for example 99, it would mean that the subject was an agent as well as an extended patient. This contradiction is sufficient reason to reject this analysis, and to prefer the first alternative. This means, in turn, that, despite the translation, the attribution of agency to the subject of this sentence is an implicature only. There is a structure in which ketinggalan does seem to have tinggalkan as the base verb. It is possible to specify the agent as the second argument: 137 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 103. Sebagian barangnya ketinggalan olehnya di hotel part goods.3 ke.leave.an by.3 hotel LOC 'He had (unintentionally) left part of his luggage at the hotel.' (ALKV: ex39) In this case, the causative verb is the only possible base, with deletion of the derivational morphology as argued for previously. It remains to account for the ketinggalan seen in example 101. I argued above that this verb cannot have an affected patient in the argument structure of its base predicate. Therefore neither the intransitive tinggal nor the causative tinggalkan are possible bases. The obvious alternative is to look for an -i suffixed verb which would make this ketinggalan quite parallel to the other verbs in ALKV's group 1d (example 101 appears in the discussion of that group in the source). There is a verb tinggali, but it means 'live in, stay at' or 'bequeath', none of which seems relevant here. However, tinggalkan has a second meaning, which is 'leave (a location)': 104. Ia meninggalkan negerinya country.3 3SG AS.leave.APPL? 'He left his country.' (E&S: 578) The questions as to how exactly this verb is derived, and whether the -kan suffix in this case is an applicative are beyond the scope of this work. We only note that there are issues to be explored here. For present purposes, it is sufficient to note that this use of the verb tinggalkan seems to be the base which we are looking for. The second argument is a location rather than an affected patient, and this brings this ketinggalan into line with the other verbs with a location as second argument, and gives a plausible reading for example 83 in which the subject is (metonymically) a location: 'I was affected because the bus left my location'. This discussion does not exhaust the possibilities of ketinggalan. This verb also allows a complement clause: 105. Petani tidak ketinggalan menghadapi kesamaan farmer NEG ke.leave.an ke.same.an AS.across.APPL selera terhadap aneka komoditas sebab appetite towards variety commodity because kemampuan beli petani sudah demikian hebat ke.able.an buy farmer already like.that intense 'Farmers have not been left behind in facing the similarity of demand for a variety of commodities because the buying capacity of farmers is already so great.' (McKay) This follows from the general principles already discussed. The verb tinggalkan can take a clausal complement with object control: 106. Saya meninggalkan Ali Ali 1SG AS.leave.CAUS 'I left Ali writing those letters.' menulis AS.write 138 surat-surat letter.DUP itu that NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 As expected, when a verb with this argument structure undergoes predicate composition with ke- -an, the patient of the base becomes subject and the new verb has subject control of the complement subject function. 3.3.3 Obliques and adjunction It is true in general in Indonesian that requiring a preposition is definitional for oblique arguments, and as discussed in section 3.2.2.1, the second argument of a ke- -an verb is rarely introduced by a preposition. It was also remarked in that section that these second arguments had properties that were reminiscent of those of actor DPs in undergoer subject clauses. Section 2.5.4 assumed rather stipulatively that an actor DP in a clause with a verb with the undergoer prefix di- was an oblique argument, even though the preposition oleh was not always used to introduce it. We can now re-examine this case, and in comparing it with the case of ke- -an verbs, offer a more complete account. There are three significant similarities between the two cases. Firstly, in order to appear without a preposition the DP must be adjacent to the verb. This fact about word order, particularly in relation to ditransitive verbs was the essential motivation for the suggestion in section 2.5.4 that such DPs are adjoined to the verb. Secondly, in the di-V clause type, it is always possible to include the preposition oleh 'by', which marks agency20. And in ke- -an clauses the preposition oleh can also introduce the second argument in some cases. This is rather rare in this clause type, but, if we consider where it is possible and where it is not, the explanation is straightforward. The discussion in ALKV (pp524-525) and the examples given in section 3.2.2.1 make it clear that the preposition can be used whenever the second argument is treated as an actor. Note that it is necessary to use the concept actor here rather than the more restrictive agent, as verbs such as kelihatan 'be visible' allow oleh with their second argument. But this is not surprising, since in its transitive form, a verb such as lihat can use oleh to mark the actor in undergoer subject clauses: 107. Pak Bomo dilihat kemarin oleh yesterday by Mr Bomo US.see 'Mr Bomo was seen yesterday by Ali.' Ali Ali The surprising fact about the use of oleh with ke- -an verbs is that it cannot be used to introduce the second argument of the verbs which SSAVI identifies as having an agent as second argument. SSAVI notes this fact as odd: that nonagentive base predicates such as lihat 'see' and tahu 'know' derive verbs which allow oleh, but the apparently agentive masuk 'enter' does not: It is very clear in Indonesian that this is the primary function of the preposition oleh, which does not have the locational meaning which by has in English. Its only other meaning is 'because'; this is not semantically unrelated and is available in rather different structures, e.g. 20 (i) Aku merasa kesal oleh AS.feel angry because 1SG 'I felt angry at his attitude.' sikapnya attitude.3 139 itu that NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 108. Setan masuk ke Mas devil enter in brother 'The devil went into Begong.' Bengong Bengong 109. Mas Bengong kemasukan (*oleh) setan brother Bengong ke.enter.an by devil 'Bengong got possessed by a devil.' (SSAVI: ex6) The undergoer subject form of the verb masuki, which I claim is the base predicate of kemasukan is perfectly grammatical with oleh: 110. Mas Bengong dimasuki oleh brother Bengong US.enter.APPL by 'Bengong was possessed by a devil.' setan devil This problem is left to future research, and for the present I will assume that oleh is possible with any ke- -an verb whose second argument is an agent. The remainder of the ke- -an verbs with two arguments have a patient or a theme as their second argument, and in these cases there is no alternative realisation with a preposition. But, across languages, the semantic notion affected patient is consistently associated with a structural position, (direct) object (Tenny 1994) to the extent that it would be surprising to find a preposition which coded those semantics. Certainly Indonesian does not have such a preposition. Therefore, for the majority of two argument ke- -an verbs the means for coding the second argument in a prepositional phrase do not exist. The third similarity between actor DPs in undergoer subject clauses and the second arguments of ke- -an verbs is in the way in which pronouns are treated in the two environments. In both cases, a pronoun adjacent to the verb is dispreferred. Where the prepositional realisation is available for the second argument of a ke- -an verb, it is compulsory for pronouns. And elsewhere, pronouns are very rare. The only verb which the native speakers I consulted would accept with a pronoun as second argument was keduluan 'be preceded by': 111. Saya keduluan dia ke.precede.an 3SG 1SG 'I was preceded by him.' Similarly, pronouns are generally impossible as actor in a di-V clause, except in the case of mereka which is argued to be an anomaly in section 2.5.3 and where they are strongly focussed, but they are possible with the preposition oleh. This is certainly true for a third person pronoun: 112. Saya dijemput oleh dia 1SG by 3SG US.meet 'I was met by him.' (IRG: 248) /olehnya by.3 First and second person pronouns are generally unacceptable as actor in a diV clause, but they are judged to be more acceptable after the preposition oleh than immediately following the verb. Thus IRG (249 n2) mentions the possibility of a phrase such as oleh saya appearing in a di-V clause, but not 140 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 the alternative. In the context of his discussion, this omission clearly means that such structures are not possible at all. The evidence discussed here makes it clear that whatever syntactic process accounts for the behaviour of actor DPs in di-V clauses should also cover the behaviour of the second arguments of ke- -an clauses. The account offered in section 2.5.4 allowed the adjunction of an actor DP, which was an oblique, to a particular morphological class of verbs. These restrictions can now be seen to be too constrained: another class of verbs allows adjunction and arguments with a variety of semantic roles can adjoin. However the restriction of the adjunction process to oblique arguments can be maintained. Adjunction to the verb is not possible in the case of non-terms which are not semantically selected by the predicate, that is, adjuncts. There is no possible alternation between structures such as examples 113 and 114 parallel to that shown in example 34: 113. Siti pergi ke Surabaya Siti go to Surabaya 'Siti went to Surabaya.' 114. *Siti Siti pergi go Surabaya Surabaya The correct generalisation then is that an oblique argument which is assigned a semantic role by a verb can be realised by adjunction to that verb. Rule 177 of chapter 2 was the following: 177. Actor adjunction (Indonesian) V → di-V DPactor ↑=↓ ↑ OBLθ = ↓ This can now be stated more generally: 115. Oblique adjunction (Indonesian) V→ V DP ↑=↓ ↑ OBLθ = ↓ Subsequent discussion in chapter 4 and chapter 5 will show that this statement is too general, but the statement is now less stipulative than that offered in chapter 2. The majority of two-argument ke- -an verbs are unusual in that their second argument is a patient and no prepositional encoding of this role is possible. Therefore their only possible realisation is adjoined to the verb. The general impossibility of pronouns as actors in di-V clauses was attributed to a constraint barring non-term pronoun actors in chapter 2. This is not a factor in the distribution of pronoun second arguments for ke- -an verbs, and this fact raises the question of whether the account in chapter 2 is correct, or incomplete. The fact that pronouns with oleh are much better following ke- -an verbs than in di-V clauses suggests that the similarities may be coincidental, and the fact that at least one ke- -an verb can occur with a pronoun adjoined to it would seem to confirm this. It is however possible that 141 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 3 adjunction structures including pronouns are avoided. A restriction of this type might be related to a difference in referentiality between pronouns and other nominals. However, I argued in section 2.3.1.2 that Myhill's (1988) account of the actor in di-V clauses as an incorporated noun fails, in part, because such actors are fully referential. It is therefore hard to see exactly how any such constraint would work. Such a constraint would explain the behaviour of the ke- -an verbs, but would not undermine the account of di-V clauses given in chapter 2, although the effects observed there might then be overdetermined by the grammar. 3.4 Conclusion This chapter has presented an analysis of ke- -an verbs in Indonesian, often described as adversative verbs. I have argued that these verbs are derived by a lexical process of predicate composition, and that the argument structure of the derived verb is organised on unusual semantic principles. The predicate ke- -an selects the most affected entity in a situation as its subject. The notion of 'most affected entity' includes a human in a close relation with the affected patient of the base verb. Examination of the syntax of clauses with ke- -an verbs has shown that many of these verbs have a non-subject argument which is an oblique, but which does not necessarily have the typical coding properties of an oblique. That is, this argument is not typically introduced by a preposition. This conclusion has forced us to extend the idea introduced in section 2.5.4, that Indonesian allows the actor in an undergoer subject clause to be adjoined to the verb. The second argument of many ke- -an verbs has similar properties to those of the actors discussed in chapter 2: they cannot be separated from the verb and they can, in some cases, alternate with expression as a prepositional phrase. Where this alternative form of expression is not possible, I argued that this is due to the lack of a semantically appropriate preposition. In the following chapter, I examine a case where a group of verbs alternate between having a prepositional phrase and a noun phrase as their non-subject arguments. But in this case, I will argue that the noun phrase is not adjoined to the verb, but is in fact a direct argument even though the verbs are not a part of the transitive system analysed in chapter 2. 142 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 4 Emotion/Cognition Predicates This chapter describes and analyses a class of predicates in Indonesian which are semantically associated with two arguments, but which do not behave syntactically in the same fashion as the verbs described in Chapter 2. Almost all of the predicates in this class denote emotional or cognitive states, a class of predicates which cross-linguistically often stand apart from the prototypical transitive pattern of a language (Belletti & Rizzi 1988, Grimshaw 1990, Kachru 1990). As the whole class of predicates has received comparatively little attention in the literature (but see Kana 1986: 283–289 and Stevens (1970)), section 4.1 argues that membership of the class in Indonesian depends on the predicate taking its second argument (the stimulus of emotion or cognition) as a prepositional phrase, and provides a description of the properties and patterns of use of these predicates. A subclass of emotion predicates also allow the stimulus to be a DP in a construction described by Stevens (1970) as 'pseudo-transitive', and this group will be the focus of investigation in section 4.2, which attempts to answer the question: what grammatical function is assigned to the second argument of such clauses? Section 4.3 recasts the analysis of section 4.2 in terms of LFG and section 4.4 discusses some theoretical and empirical issues which arise from this analysis. 4.1 A class of predicates 4.1.1 Criterion of membership I stated above that I take the property of coding the second argument as a PP to be the appropriate defining property for the class of predicates under consideration: 1. Sampai Retno lupa pada dirinya until Retno forget to self.3 'Until Retno forgot herself.' (PYD: 187) 2. Jadi then sang rakyat boleh mual dan bosan dengan populace may feel.queasy and bored with PM PM HON 'Then the people may feel dissatisfied and bored with the honourable PM.' (McKay) Both Stevens (1970) and Kana (1986), while acknowledging the existence of the prepositional construction, assume that appearance of a predicate in the 'pseudo-transitive' construction is criterial. There are two reasons for rejecting this approach. Firstly, there are Indonesian predicates which are clearly of the same semantic type as those considered by Stevens and Kana but which cannot occur in the 'pseudo-transitive' construction. For these predicates, the prepositional construction is obligatory, not optional. An example is the predicate puas 'satisfied, content': 143 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 3. Dan mereka juga sudah puas dengan kasus and 3PL also content with case PERF wanita yang hilang itu? woman REL lost that ‘And were they also satisfied with the case of the woman who disappeared?’ (SDPGS: 128) 4. *Dan mereka juga sudah puas kasus wanita yang hilang itu? Other predicates which behave like this are cinta 'love' and setia 'loyal': 5. Aku tetap cinta 1SG constant love 'I still love you.' (E&S:118) padamu to.2 6. *Aku tetap cinta kamu. 7. kakek-kakek yang setia padaku grandfather .DUP loyal to.1SG REL 'old relatives who are loyal to me.' (PYD:92) 8. *kakek-kakek yang setia aku (but OK as 'my loyal old relatives' In other respects, such predicates share many properties with the 'pseudotransitives': they typically derive verbs which participate in the system analysed in chapter 2 when an applicative suffix is added: 9. Kurasa dia mencintaimu 1SG.feel 3SG AS.love.APPL.2 'I feel that he loves you.' (PYD:39) And some of them can take a non-finite clause as complement, often with a controlled subject position: 10. Sedang kita sendiri hanya puas menjadi only content become while 1PL.INCL INT.self pengikut-pengikutnya follower.DUP.3 'While only we ourselves are content to become his followers.' (McKay) All of these properties are discussed in more detail in following sections. My point here is that there is a clear overlap in properties between predicates which only ever occur in the prepositional construction and the 'pseudotransitive' predicates. The most satisfactory approach therefore is to treat the 'pseudo-transitives' as a subset of the more general class. The second reason for preferring this approach is that the prepositional construction is widespread across the Austronesian language family. Many languages in the Western Austronesian group have this feature, but the 'pseudo-transitive' construction is less common. Examples of such languages are Acehnese, Balinese, Sasak and Tagalog. 144 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Acehnese has many verbs which occur in the prepositional construction, and which are formally intransitive. This language has a classic fluid-S system (Dixon 1994:78–83). Transitive verbs have obligatory proclitics crossreferencing the actor; the undergoer is optionally cross-referenced by an enclitic1. Some intransitive verbs allow either type of coding for their argument, with a corresponding semantic difference and this group includes some emotion verbs2: 11. gopnyan hanainseuehgeuh keu- lôn he/she feel compassion 3 1 NEG-IN DAT 'He/she has no sympathy towards me.' (Durie 1985:4-13:49) 12. gopnyan he/she hanaNEG-IN geu3 inseueh feel compassion keuDAT lôn 1 'He/she feels no sympathy towards me (by choice).' (Durie:1985:420:49) Example 11 makes it clear that this verb is intransitive: otherwise a proclitic would be required. Durie suggests that the preposition keu is omissible (1985:61), and gives an example: 13. h'an- galak- geuh (keu)- umuda like 3 (DAT) coconut young NEG 'He doesn't like young coconut.' (Durie 1985: 4-83, 62) My own research suggests that the preposition is not generally omissible, and is obligatory when the following DP is animate. However, the speaker with whom I worked spoke a different dialect to that described by Durie, and this variation might be due to this difference. Acehnese shares the prepositional construction and the 'pseudo-transitive' construction with Indonesian, and there is some evidence that the prepositional construction is basic. Balinese also has semantically similar verbs which introduce the stimulus argument with a preposition, and Artawa, Artini and Blake (1997) refer to these verbs as 'two-place intransitives'. They include tresna 'to love', inget 'to remember' and demen 'to like': 14. cerik-cerik children jani now lebihan more sing NEG demen like teken with jaja roti cake Children today more and more do not like cakes (Artawa, Artini & Blake 1997:4 ex8) The sources which I have consulted do not give any examples of 'pseudotransitive' constructions, but it is possible to make the stimulus a direct argument of the verb, and the verbal morphology shows that its argument structure is changed to allow this3: 1 Some dialects lack enclitics altogether (Bukhari Daud, p.c.). 2 Glossing as in the source: IN = inchoative. 3 Glossing as in source: DEF = definite, AD = advancement. 145 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 15. Tiang 1SG demen like gati much ajak with rok-e dress-DEF 'I like the dress.' 16. Rok-e dress-DEF demen-in like-AD tiang 1SG 'I like the dress.' (Artawa, Artini & Blake 1997:13 ex32) The additional verbal morpheme in example 16 which Artawa et al. gloss as AD(vancement) is clearly parallel to the Indonesian applicative in example 9 above, and also probably cognate to it (see Artawa 1999: 43-45). Note also that when the verb becomes transitive in example 16, the actor appears in the postverbal position required for a Balinese transitive verb without a nasal prefix (Artawa and Blake 1997). On the available data, it seems clear that the prepositional construction can be considered basic in Balinese. Tagalog also has a group of verbs which behave in a similar fashion. In the simplest case, the experiencer argument is marked with the nominative case marker ang, and the stimulus is marked by the oblique case marker sa: 17. na-gulat AS-surprised ang NOM babae woman sa OBL boyfriend boyfriend niya 3SG.POSS 'The woman was surprised at her boyfriend.' (Martin 1996: 252 ex39) The case marking possibilities of Tagalog mean that the language does not allow the 'pseudo-transitive' construction. And if a non-core argument is promoted, it must become subject. Therefore the only alternative construction is one in which the stimulus is subject while the experiencer is still a core argument (i.e. a transitive construction): 18. k-in-a-gulat-an US-surprised ng CORE babae woman ang NOM boyfriend boyfriend niya 3SG.POSS 'The woman was surprised at her boyfriend.' (Martin 1996:252 ex40) The verbs which behave in this fashion are semantically similar to those in Indonesian and Balinese. Additional evidence as to the spread of this construction within the Austronesian family is not hard to find. For example, the so-called 'middle verbs' of Polynesian languages have much in common with this construction (Chung 1978b: 216-234, Seiter 1978). A very small number of predicates do not fit the general pattern. These predicates do not have transitive forms (except with applicative suffixes), but they also never occur in the prepositional construction. The predicates in this group include mau 'want', punya 'own, possess', tahu 'know' and perlu 'need'. A further noteworthy feature of this group is that these verbs seem able to become auxiliaries very easily. Note the position of mau relative to the actor pronoun in the following example: 146 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 19. Nasi itu mau dia makan rice that want 3SG eat 'That rice, he wants to eat it.' (Waruno Mahdi, p.c.) In many cases, mau has an almost purely aspectual meaning and can occur with non-sentient subjects: 20. Rumah itu mau roboh house that want collapse 'The house is about to collapse.' (Waruno Mahdi, p.c.) The predicate ingin 'want, desire' should probably also be included with this group. 21. Ingin dia makan, nasi itu eat rice that wish 3SG 'He wishes to eat it, that rice.' (Waruno Mahdi, p.c.) Kana (1986: 284) has an example of this predicate in the prepositional construction, but I have no such examples in my data and E&S does not give such an example. It may be significant that the preposition in Kana's example is akan; this preposition is rarely encountered in most registers of contemporary Indonesian (see discussion in section 4.1.2.2 below) and has a distinctly formal or archaic feel. I would therefore suggest that ingin may previously have behaved in the same way as the main group of emotion predicates, but that it is moving (or has moved) to the anomalous group. Quite plausibly, the other predicates in this anomalous group have followed a similar path. Several of these predicates can also be used as auxiliaries, but lacking the prepositional construction is neither a necessary nor a sufficient condition for being used as an auxiliary. On the one hand, punya and tahu are not used as auxiliaries, and on the other hand suka, which retains the prepositional construction, is (see further discussion in section 4.1.2.7 below). 4.1.2 Range of constructions Most of the predicates under consideration here can appear in a range of constructions. In addition to the prepositional construction and the 'pseudo-transitive' construction, most predicates form a transitive verb with an applicative suffix, and a small group form transitive verbs directly without an applicative. Transitive verbs are also derived from some predicates with the causative suffix -kan. Many predicates can also take a complement clause as an argument. There are various possibilities in this regard: a full clause with a complementizer, a full clause with no complementizer, or a reduced clause with the subject position controlled. Finally, non-predicative uses are possible for some of these words: some appear as nouns and some as nominal modifiers4. The following sections give examples and brief discussion of each of these constructions in turn. For each predicate for which I have a textual The question of to what lexical category the words being investigated should be assigned is discussed in detail in section 4.1.3. 4 147 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Word bangga benci1 bimbang bosan cemas cemburu cinta curiga ganggu gemar gemas gila heran ingat2 jemu jengkel kangen kasih kasihan kecut kenal3 kuatir lupa malu marah mimpi mirip peduli percaya4 prihatin puas sadar salut sayang senang setia simpati suka takut5 tega yakin Gloss proud hate anxious bored worried jealous love suspect upset fond of annoyed insane surprised remember fed up annoyed long for love pity afraid know s.o. afraid forget ashamed angry dream resemble care about believe concerned satisfied aware respect pity, love like loyal sympathetic like afraid have the heart to sure DP + Applic (-kan) Caus + -kan + + + (-i) -i -i + + + Ø+S + + (-i) (-i) -i -kan -kan + + + + -i (-i) -kan -kan -kan -i -kan (-i) + + + + + + + + + + + + (+) + + (+) + + + + (+) + + + (+) + Control + + + + (+) (-i) -i -i + + + + + C+S + + + + + + + + + + + + + + -i/-kan -i (-i) + + + + + + -i -i/-kan + + + + + + + + + + -i + + + + Table 4.1 – Indonesian Emotion and Cognition Words Occurrence in various clause types 148 + + NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Table 4.1 – Legend and notes DP Can have a bare DP as complement Applic Takes the specified suffix to derive a transitive verb Caus Takes the suffix -kan to derive a causative verb C+S Can be followed by a full clause following a complementizer Ø+S Can be followed by a full clause without a complementizer Control Can be followed by a clause with a controlled subject position A plus sign in any column indicates that at least one textual example of the clause type has been recorded. A bracketed plus sign indicates that native speakers believe the clause type is possible, but that no textual examples have been recorded. In the column Applic, the entries indicate which applicative suffix, if any, is used with the word. Bracketing has the same meaning as in all the other columns: a native speaker attests the usage but no text example has been collected. Notes: For discussion of whether benci has an applicative formation with -i, see section 4.1.2.4. 1 There is a transitive verb derived directly from ingat without an applicative suffix. Therefore the examples with an DP complement are dubious. 2 There is a transitive verb derived directly from kenal without an applicative suffix. Therefore the examples with an DP complement are dubious. 3 There is a transitive verb derived directly from percaya without an applicative suffix. Therefore the examples with an DP complement are dubious. 4 The suffix -i can also derive a causative with takut. See section 4.1.2.4 for discussion. 5 149 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 example of the prepositional constructions it can appear in. construction, Table 4.1 shows which 4.1.2.1 THE DATA The data on which this survey is based are drawn from the following sources: 1) Two contemporary novels, one set in Jakarta (SDM) and one set in Surabaya (PYD). I have not attempted to analyse any variation which might be caused by this difference, as other factors are not controlled for. In particular, PYD is more 'slangy' in tone. However, the central characters in each novel are educated Indonesians, and I assume that these sources have made some attempt to accurately portray the speech of such persons. 2) One translation of an English novel into Indonesian (SDPGS). This source has certainly introduced some distortions into my data in terms of relative frequencies of certain constructions. For example, Miss Christie's characters are fond of adding the expression 'I dare say' to their comments. This is translated as Saya berani katakan..., and I am sure that the appearance of the predicate berani with reduced complement is consequently over-represented in my data. 3) Examples drawn from a corpus of contemporary Indonesian journalism collected by Helen McKay. 4) Example sentences provided by E&S. This is the standard IndonesianEnglish dictionary, and its usage is probably conservative. Most of the examples which I have using the preposition akan, which is no longer common in spoken Indonesian come from this source. 5) Native speaker judgments. The two speakers consulted are both educated Indonesians of Javanese background. Full bibliographic details of the sources described in 1,2 and 4 are provided immediately before the reference section of this work. Pending the discussion of the lexical categorization of the words under discussion, I use the neutral term predicate throughout this section. I refer to the two semantic roles associated with these predicates as experiencer and stimulus. 4.1.2.2 PREPOSITIONAL CONSTRUCTION A wide range of prepositions occurs in this construction. At least historically the most widespread in usage is akan, which marks the second argument of all two place verbs in classical Malay (Cumming 1991: 42-45). It is therefore theoretically possible with all the emotion predicates. However, it is not commonly used in non-formal registers of Indonesian now. Occasional examples do occur 22. Retno sangat takut akan perpisahan dengan Retno very afraid about separation with 'Retno was very afraid of parting from Sandy.' (PYD: 183) 150 Sandy Sandy NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 The most common prepositions used are dengan 'with' and (ke)pada 'to, at'. A native speaker has informed me that she was taught, as a prescriptive rule, that in this construction dengan should be used with non-human nouns following, and kepada with humans. The data shows this pattern as a trend, but it is certainly not observed as an absolute rule: 23. Dan orang yang paling tidak puas dengan and person REL important NEG content with keputusan itu decision that 'And the person who is least satisfied with that decision' (SDM:28) 24. saya percaya ia tidak kenal dengan orang 1SG believe 3SG NEG know with person semacam itu certain kind that 'I don't believe he knows anyone like that.' (SDPGS: 245) 25. Barangkali ia hanya bosan pada hotel itu maybe 3SG only bored to hotel that 'Maybe he just got bored with the hotel.' (SDPGS: 101) 26. Dia sangat benci padaku. 3SG very hate to.1SG 'She really hates me.' (SDM:266) 27. Marilah kita memperkuat tekat untuk tetap 1PL.INCL AS.support resolve for constant please.EMPH setia kepada aspirasi loyal to aspiration 'Let us strengthen our commitment to remain loyal to the ideals.' (McKay) 28. Marisa sangat takut kepadanya Marisa very afraid to.3 'Marisa was very frightened of him.' (SDM:259) There is a colloquial equivalent of dengan, sama 'together with', and this also is used in this construction. All the examples I have in my database have human stimuli: 29. Aku kangen sama long for with 1SG 'I long for her.' (PYD:33) dia. 3SG The preposition tentang 'about' can occur with some predicates, typically cognitive rather than affective ones. However, the only textual examples which I have for this preposition come from the source translated from 151 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 English. Most of these appear to be direct translations of the English expression 'sure about ...': 30. Anda yakin tentang hal itu? 2SG sure about matter that 'Are you sure about that matter?' (SDPGS: 147) In that source, there are also examples with heran 'surprised, astonished' and peduli 'attend to, care about'. In each case, direct translation from English seems likely. These examples suggest that although this preposition is not ungrammatical with some cognitive predicates, the usage is probably not natural for native speakers. Table 4.2 shows the possible co-occurrences of predicates and prepositions, and the data presented there reinforce this point. The distribution of tentang is restricted in comparison to the other five prepositions listed there, suggesting again that this preposition is only used in translation of English (and perhaps Dutch) expressions. Three other prepositions occur in isolated examples: ke 'to', atas 'on'5 and terhadap 'about, toward': 31. Dia malah marah ke Presiden Peru Fujimori 3SG instead angry to President Peru Fujimori 'Instead, he was angry with Peru's President Fujimori.' (McKay) 32. Saya senang sekali atas konsolidasinya happy very on consolidation.3 1SG 'I am very happy about the merger.' (McKay) 33. organisasi yang peduli terhadap permasalahan organization REL care about set.of.problems kependudukan demography 'an organization which cares about demographic problems' (McKay) Note that all of these examples are from journalism; it is possible that contemporary usage allows a wider range of prepositions in this construction. There are two generalisations which can be made about the use of prepositions in this construction. Firstly, the general locative preposition di is never used, and the only spatial concept which plays any role in this construction is movement towards. More recent usage may be altering this picture, with the appearance of atas and terhadap which can mean 'across from' or 'opposite'. This fact will be discussed further in section 4.2 in relation to Kana's (1986) claim that the stimulus in this construction is a Location. This is a fact about Indonesian; other Austronesian languages use locative prepositions in their parallel construction. For example, in Sasak (Menó-mené variety), the locative preposition léq is used with emotion predicates: Strictly speaking, atas is a locational noun and it normally occurs together with the general locative preposition di, thus di atas 'on top of'. By a transparent reanalysis, atas has come to be used as a preposition in its own right. A similar process can be seen with dalam 'inside', from di dalam 'at the inside of'. 5 152 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Word bangga benci bimbang bosan cemas cemburu cinta curiga ganggu gemar gemas gila heran ingat jemu jengkel kangen kasih kasihan kenal kuatir lupa malu marah mimpi mirip peduli Gloss proud hate anxious bored worried jealous love suspect upset fond of annoyed insane surprised remember fed up annoyed long for love pity know s.o. afraid forget ashamed angry dream resemble care about akan + + + (+) + (+) (+) + (+) + (+) (+) (+) + ? (+) (+) (+) (+) (+) (+) (+) (+) (+) (+) (+) + dengan + + ? + (+) (+) (+) (+) + (+) (+) + (+) + + + (+) (+) (+) + (+) + (+) + percaya prihatin puas sadar salut sayang senang believe concerned satisfied aware respect pity, love like + + (+) + + (+) + + (+) + (+) + (+) + setia simpati suka takut5 tega loyal sympathetic like afraid have the heart to sure (+) (+) (+) + (+) (+) yakin + + + + + + kepada + + ? (+) (+) + (+) (+) pada (+) + ? + (+) + + + sama (+) (+) ? (+) (+) other + (+) terhadap (+) (+) ? (+) (+) (+) (+) (+) (+) (+) (+) (+) (+) (+) + + + ? (+) (+) + + + (+) + (+) + (+) (+) (+) + ? (+) + (+) (+) (+) (+) (+) + (+) (+) + (+) (+) + (+) (+) + (+) + (+) (+) (+) (+) + + (+) (+) (+) (+) (+) + (+) + (+) + + (+) + + + + + (+) (+) (+) (+) (+) (+) (+) (+) Table 4.2 – Indonesian Emotion and Cognition Words Co-occurrence with various prepositions 153 atas tentang tentang (tentang) di ke (tentang) tentang, terhadap atas atas (tentang) atas terhadap terhadap tentang NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 34. Takutk léq acòng afraid.1SG LOC dog 'I am afraid of dogs.' This fact also leads to the second generalisation, that the semantic contribution of the preposition is not important. This second generalisation is illustrated by Table 4.2 which shows the co-occurrence possibilities for each predicate in the database and six prepositions. Five of the prepositions (tentang being the exception) can each occur with almost every predicate, the gaps which do appear seeming to be quite idiosyncratic. For many predicates, examples occur with more than one preposition, and in one case, there are examples with five prepositions (see also example 33 above): 35. akan Saya tidak peduli perkataannya care about 1SG NEG NOM.word.NOM.3 'I don't care what he said.' (E&S: 415) 36. tentang Mereka tidak peduli sedikit pun 3PL care a.little about NEG EMPH 'They didn't care even a little about him.' (SDPGS: 184) 37. peduli dengan Moskow seolah tak segala macam Moscow as.if care with all type NEG reaksi di dunia. reaction LOC world 'Moscow apparently doesn't care about any sort of reaction from the world.' (McKay) 38. peduli pada Mereka seperti tak Kepala 3PL as care to head NEG Bulog Beddu Amang distribution.agency Beddu Amang 'They don't seem to care about the chief of the distribution agency, Beddu Amang.' (McKay) dia 3SG Such evidence suggests very strongly that there is no semantic dependency between the preposition and its complement in this construction, and that the stimulus is assigned its semantic role directly by the predicate. If a predicate in the database occurs with more than one preposition, the two used are most likely to be dengan and (ke)pada. There is no common semantic ground between these two, and they are not mutually substitutable in other constructions. (ke)pada is the preposition normally used for the recipient in act of giving, but dengan is impossible in such a context: 39. Ia memberikan uangnya kepada adiknya 3SG AS.give.APPL money.3 to brother.3 'He handed over his money to his younger brother.' (E&S: 73) 154 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 40. Ia memberikan uangnya dengan adiknya 3SG AS.give.APPL money.3 with brother.3 'He handed over his money with(/*to) his younger brother.' dengan is typically used to indicate a co-participant in an action, and pada cannot have this meaning either: 41. Ia pergi dengan ayahnya with father.3 3SG go 'She went with her father.' (E&S: 138) 42. Ia pergi pada ayahnya 3SG go to father.3 'She went to(/*with) her father.' As the above examples also show, in other contexts there is no basis for the tendency (and prescriptive rule) noted above for dengan to occur with inanimate stimuli. In English, some predicative adjectives display similar behaviour: 43. John was angry with her. 44. John was angry at her. The tendency for a distinction between the prepositions appropriate for human stimuli and those appropriate for non-humans is also seen (at least to some extent) in English: 45. Mary is annoyed with/?about Bill. 46. Mary is annoyed about/?with the traffic Example 43 is possible with the preposition about, but the most natural reading is that Mary is annoyed about some action preformed by Bill, not by Bill himself. The use of with in example 44 seems less felicitous, and creates an anthropomorphising effect. The general principle is that when a predicate semantically selects a complement without licensing it as a direct argument, then the licensing element must be semantically compatible with the selected role. In the case where a language does not have a licenser, such as an adposition or casemarker, which is semantically specific for the role selected, any other linker which is compatible can be used. Thus, neither English nor Indonesian has a preposition which specifically codes the semantic role 'stimulus of emotion', and predicates which select this role without licensing it can use any other preposition which is compatible (and this does not include purely locational ones). Where a preposition is specific, it is the obligatory linker as for agents in both languages (by in English, oleh in Indonesian). The intermediate case is a verb like put which semantically selects a location as one of its arguments without licensing it directly. In consequence, any compatible licenser is acceptable. 155 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 A small number of predicates occur with one of the intransitive verb prefixes ter- and ber- in this prepositional construction. The ter- prefix usually has a meaning of involuntary activity, typically translated by an English passive: 47. Ardi teringat pada wajah seekor. Ardi ter.remember to face CLASS(animal) 'Ardi was reminded of the face of an animal.' (SDM: 16) 48. Saya berpikir tentang ber.think about 1SG 'I think about Java.' pulau island Jawa. Java These examples are not surprising: these two prefixes are generally agreed to occur with intransitive verbs and they therefore fall under the same general pattern as that discussed thus far in this section. What is surprising, is that occasional examples appear of transitive verbs with a following preposition: 49. Aku sendiri tak mengerti 1SG self NEG AS.understand 'I myself don't understand you.' (PYD: 194) denganmu with.2 which can be compared with a more normal usage for this verb: 50. Ia mengerti maksud surat itu 3SG AS.understand intention letter that 'She understands the meaning of that letter.' (E&S: 160) This verb is clearly in the same semantic field as the predicates under discussion, and it would seem that the prepositional construction exerts an influence. Occasional examples also occur in which the predicate is derived as a transitive verb with an applicative suffix, but a preposition also appears before the stimulus: 51. Yang penting saya sangat mencintai Sandy important 1SG very Sandy AS.love.APPL dan menyayangkan atas semua kejadian itu and AS.pity.APPL on all event that 'What is important is that I love Sandy very much and that I regret all these events.' (PYD: 200) REL The issue of where applicative verbs in Indonesian can or should be followed by a preposition is complex (see Chung 1976a for some discussion). The confusing cases seem to be limited to three place predicates, where the third argument may have a preposition even when a derived verb is used. Derived two-place verbs generally behave as straightforward transitive verbs, and the above example (51) is atypical. 4.1.2.3 BARE NOUN-PHRASE COMPLEMENTS Many of the emotion and cognition words which appear with a prepositional phrase complement can also occur with a bare DP complement. 156 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 This clause type is discussed in detail in section 4.2, therefore I only provide examples here to illustrate the construction: 52. rumahku? Kamu lupa 2SG forget house.1SG ‘You have forgotten my house?’ (SDM: 96) 53. konduite jelek Mau dijadikan uang, saya takut want US.cause money 1SG afraid report bad 'Wanting money to be made, I am afraid of a bad report.' (McKay) 54. Cuma aku tidak suka omongannya yang sok only 1SG like word.3 as.if NEG REL 'I just don't like his speeches that take the moral high ground.' (PYD: 87) 55. Saya nggak senang wartawannya like journalist.3 1SG NEG 'I don't like the journalist.' (McKay) suci, pure Not surprisingly, many clauses of this type are like the examples just given in having 1st or 2nd person pronouns as experiencer. But other types of nominal are possible in this semantic role, including non-pronominal 3rd persons: 56. Dua manusia sesama jenis bercumbuan di pojok two human equal kind flirt corner LOC jalan, tanpa peduli keramaian sekitar road without care around NOM.busy.NOM ‘Two human beings of the same sex flirt at the street corner, without caring about the surrounding bustle.’ (McKay) 57. Kalau boleh memilih, Ardi lebih suka Marisa if permitted AS.choose Ardi more like Marisa bersikap selembut penampilannya ber.attitude as.soft appearance.3 ‘If he had a choice, Ardi would have preferred Marisa to be as gentle as her appearance.’ (SDM : 27) This clause type is therefore not a type of Pro-V clause with non-standard word order. Speakers report that there is no semantic difference between the clause type with a preposition, and that which lacks it. The following examples are adjacent clauses from dialogue: 58. Papa sayang padamu, father love to.2 'Father loves you, Risa'. Risa. Risa 157 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 59. Risa juga sayang Papa. Risa also love father 'Risa loves Father too.' (SDM: 293) This is the moment in the story at which the heroine is reconciled with her step-father, and the author’s intention is to portray a mutuality of emotion. Even in this context it is acceptable for one character to use one clause type and for the second character to respond with the other type. In section 4.2, I argue that the DP complement in clauses such as example 59 are direct arguments, while the PP in examples like 58 contains an oblique argument. The lack of semantic differentiation is therefore somewhat surprising. Where there is a choice between coding an argument as an oblique or as a direct argument, the direct argument is usually more fully implicated in the action. A well-known example of this phenomenon in English is the conative alternation: 60. The hunter shot the deer. 61. The hunter shot at the deer. where example 60 entails that the bullet hit its target, whereas example 61 does not. Similar remarks apply to the argument alternation possible with verbs of the spray/load class (Hall 1965, see also Dowty 1991: 587 and references cited there). The Indonesian case is different according to the evidence presented here, and I would suggest that this difference is linked to the fact that, as discussed in the previous section, the preposition makes no semantic contribution. This question is discussed further in section 4.1.3.2. 4.1.2.4 APPLICATIVES AND CAUSATIVES The majority of the emotion/cognition predicates derive transitive verbs with one of the applicative suffixes -i and -kan (-kan also derives causatives, discussed later in this section). Typical examples for each type are: 62. Kurasa dia mencintaimu 1SG.feel 3SG AS.love.APPL.2 'I feel that she loves you.' (PYD: 39) 63. Aku 1SG itu, that 'I am 195) melupakan senang kau sudah bisa wanita happy 2SG able AS.forget.APPL woman PERF San Sandy happy that you were able to forget that woman, Sandy.' (PYD: Kana (1986: 287) implies that lupa is the only predicate which takes the -kan suffix in this construction, but examination of Table 4.1 shows that this is not the case. Applicative derivation results in a verb which can participate fully in the system described in Chapter 2. Both verb prefixes are possible and are used to make arguments accessible for processes such as relativisation: 158 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 64. Kamu punya ibu yang menyayangimu. 2SG have mother REL AS.love.APPL.2SG 'You have a mother who loves you.' (SDM: 84) 65. Untuk menemui anak yang begitu disayanginya. for child so AS.meet.APPL REL US.love.APPL.3 'In order to meet a child who is so loved by her.' (SDM: 208) The pronoun agent construction is possible; note the relative position of the pronouns and the negation in example 66 and auxiliary in example 67: 66. kau sukai Apakah ada sesuatu yang tak what.EMPH be something REL 2SG like.APPL NEG tentang Nicky? about Nicky 'Is there something you don't like about Nicky?' (PYD: 82) 67. Anda Itulah yang telah lupakan 2SG forget.APPL that.EMPH REL PERF 'That is what you have forgotten.' (SDPGS: 314) The bare verb actor subject construction is also possible; note the position of the adverb sangat immediately before the verb sayangi in the following example: 68. Yakni, bahwa salah seorang dari yang hilang that.is that one one.person from REL missing itu adalah seorang yang kami dari divisi that COP one.person REL 1PL.EXCL from division sayangi sangat kenal, sangat very know very love.APPL 'That is, the fact that one of those who were missing is a person who we - from the division - know so well, love so much.' (Simatupang 1993: 117, quoted in Voskuil 1996: 212) The data examined here gives little support to theories which differentiate the two applicative suffixes semantically (Tampubolon 1983, Kana 1986, Voskuil 1996). Although there is a tendency for predicates with more cognitive meanings to take the -kan suffix and for more emotional meanings to occur with the -i suffix, the division is not clear cut. For example, it is not obvious why curiga 'suspect' and percaya should behave more like emotion words than bangga 'proud' or peduli 'attend to, care about'. There are also at least two predicates which can derive transitive verbs with both applicative suffixes. One is sayang 'pity, love' which has the derivatives sayangi 'love' and sayangkan 'regret, deplore'. Although the two derived verbs have different meanings, the thematic relations between the verb and its arguments are essentially the same in each case. Even more complex is the case of takut 'fear' which derives transitives with both suffixes, with both derivatives also having a causative reading: 159 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 kita 1PL.INCL takuti fear.APPL 69. Hanya Tuhan yang sewajarnya only God truly REL 'We need fear only God.' (E&S: 544) 70. Tapi yang ditakutkan oleh Mustika bukanlah wanita but by Mustika NEG.EMPH woman REL US.fear.APPL yang sedang dikejar-kejar oleh Sandy by Sandy REL PROG US.chase.DUP 'But what Mustika feared was not the woman who was being chased by Sandy...' (PYD: 30) 71. menakuti Ia dan membentak orang bawahannya 3SG AS.fear.CAUS and AS.snap.at person subordinate.3 'She intimidated her subordinates and spoke harshly to them.' (E&S: 544) 72. berarti ia melangkah ke dalam bahaya yang in inside danger meaning 3SG AS.step REL menakutkan AS.fear.CAUS 'meaning that he stepped into a terrifying danger.' (SDPGS: 274) It is not common to find the -i suffix used as a causative, but this data show that it is possible. These particular examples raise a problem in that the same verb forms can have the experiencer linked to either grammatical function, suggesting that two lexical entries would be required for each verb form. As shown in Table 4.1, causative derivations are possible with many predicates, of both the emotion and cognition types: menjengkelkan lebih lagi, mertuanya ikut more AS.angry.CAUS again parent-in-law follow REL campur tangan. mixed hand 'What was more annoying, her in-laws were involved.' (SDM: 110) 73. Yang 74. Ia selalu menasehati 3SG always AS.advice.APPL dirinya berhati-hati self.3 careful dan and mengingatkan AS.remember.CAUS 'He always advised and reminded himself to be careful.' (PYD: 167) Indonesian6 systematically lacks lexical causatives in this semantic field. English verbs of this type such as remind (example 74) and frighten (example 72) are always translated by derived Indonesian verbs. A possible restriction on the use of the -kan suffix as a causative is discussed further in section 4.3.1.2. 6 And other Austronesian languages, as far as my research has gone. 160 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 4.1.2.5 TRANSITIVE VERBS A few predicates can form transitive verbs without a derivational suffix. The most common of these is ingat 'remember': 75. Sandy masih melamun mengingat kehangatan Sandy still warmth AS.muse AS.remember tubuh Nicky body Nicky 'Sandy still daydreamed remembering the warmth of Nicky's body.' (PYD: 142) The other common form is benci 'hate': 76. Salahkah kalau dia membenciku? mistake.EMPH if 3SG AS.hate.1SG 'Is it wrong if he hates me?' (SDM: 229) Some native speakers claim that in such clauses benci is an applicative verb with -i, but there is no evidence that this is true, either synchronically or diachronically (Sander Adelaar, p.c.). If a predicate forms a simple transitive verb, clauses in which the bare form occurs with a bare DP stimulus cannot be counted as examples of the construction considered in section 4.2. The clause could be a transitive clause with a bare verb form in such cases, and I do not include verbs with simple transitive forms as participating in the DP stimulus construction. 4.1.2.6 INTRANSITIVE USES Most of the predicates under discussion can be used in a clause with one argument: 77. Ia tetap menjadi suami yang 3SG constant AS.become husband REL 'He certainly became a loyal husband.' (SDPGS: 70) 78. Pasti ia pergi karena takut because afraid certainly 3SG go 'Certainly he left because he was afraid.' (SDPGS: 57) setia loyal In such examples, the predicate behaves like an adjective. (The issue of what lexical category or categories these predicates should be assigned to is discussed in 4.1.3 below.) But with many predicates, an objectless emotion or cognition cannot be construed, and in such cases a second argument is clearly implied: 79. aku sudah nggak ingat lagi 1SG remember again PERF NEG 'I did not remember (it) any more.' (PYD: 60) 161 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 80. dan Mustika nampak and Mustika apparently Sandy yang nampak Sandy REL apparently 'and Mustika wanted to cling not to like (it).' (PYD: 12) manja dan bersandar ke clinging and lean to tak begitu suka like.that like NEG to and lean on Sandy who appeared Such examples show that many of these predicates do require two arguments in their semantic representation. 4.1.2.7 CONSTRUCTIONS WITH COMPLEMENT CLAUSES Many emotion and cognition predicates occur in clauses in which the second argument is a complement clause. In fact, for some of the most common predicates this is the most common type of appearance: this is true for suka 'like' and senang 'happy, like' and also for yakin 'sure', although the data studied may not be representative for this predicate (see discussion in section 4.1.2.1 above). Three subtypes of this construction occur. Firstly, the complement can be a full clause introduced by a complementizer: 81. Tetapi, but adalah akhirnya ia yakin bahwa yang dihadapinya finally 3SG sure that REL US.face.3 Kwalik Kwalik COP 'But, in the end he is sure that it is Kwalik who will face him.' (McKay) 82. dia takut kalau-kalau Retno tiba-tiba ingin ikut 3SG afraid if.DUP Retno suddenly wish follow bersamanya ke Jakarta together.3 to Jakarta 'He was afraid that Retno might suddenly want to go with him to Jakarta.' (PYD: 181) A variety of complementizers are possible in this construction. Besides the two seen in the preceding examples, the possibilities include jika 'if', untuk 'for', and various interrogative expressions such as bagaimana 'how' and mengapa 'why'. The second possibility is the same as the first, but with the complementizer omitted: 83. Yang sudah tidak peduli orang lain percaya care.about person other believe REL PERF NEG atau tidak dengan ucapannya or not with statement.3 'Who did not care whether others believed his statements or not.' (SDM: 231) 162 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 84. Aku tidak suka kau bergaul dengan Nicky 1SG like 2SG socialise with Nicky NEG dalam inside 'I don't like you to socialise too closely with Nicky.' (PYD: 82) terlalu too See Kana (1986: 241-252) for discussion of whether this type of structure should be analysed as 'raising to object'. Finally, and most commonly, the subject of the complement clause is co-referential with the subject of the matrix and is omitted, a control structure: 85. Aku masih takut jatuh cinta still afraid fall love 1SG 'I am still afraid to fall in love.' (PYD: 12) 86. Mata Sandy seakan tak percaya melihat mereka believe 3PL eye Sandy as.if NEG AS.see "It was as though Sandy's eyes could not believe that they saw them.' (PYD: 202) The predicate suka 'like' occurs very commonly in this construction: 87. Karena Tika memang gadis yang baik dan because Tika certainly girl good and REL suka menolong like AS.help 'Because Tika was certainly a good girl and liked to help.' (PYD: 61) 88. Kamu lebih suka bermain-main mencari kesenangan more like play happiness 2SG AS.seek sendiri daripada bekerja self from work 'You prefer to play around seeking your own happiness rather than working.' (SDM: 61) This word can also be used as an auxiliary in the same way that mau and ingin can (see section 4.1.1): 89. Nasi itu suka dia makan rice that like 3SG eat 'That rice, he likes to eat it.' (Waruno Mahdi, p.c.) This usage is plausibly a grammaticalization of the control structure described above, and the process has progressed to the extent that the selectional restrictions which suka exerts over its subject, which must normally be human or minimally conscious, is relaxed: 163 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 90. Bis kota itu suka terlambat bus city that like delayed 'That bus is often / usually behind schedule.' (Waruno Mahdi, p.c.) 4.1.2.8 NOMINAL USES Emotion and cognition words can occur in typically nominal environments. The most common of these is as apparent object of the verb (me)rasa(kan) 'feel': 91. supaya tidak merasa bosan in.order.that NEG bored(om) AS.feel 'in order not to feel bored(om) (SDPGS: 58) It is possible though that merasa selects an adjective as its complement, and that the emotion word here is not a noun. This interpretation is favoured by the fact that such complements are negated by tidak, which negates verbs and adjectives, rather than bukan, the usual negator for nouns: 92. Ia merasa tidak puas satisfied 3SG AS.feel NEG 'He felt unsatisfied.' (SDPGS: 229) However, emotion and cognition words can appear as the complement of other verbs in more clearly nominal contexts: 93. Pertemuan [...] itu tampaknya mengundang simpati sympathy meeting that apparently AS.invite dunia terhadap kebijaksanaan Vietnam world towards policy Vietnam 'The meeting has apparently brought the sympathy of the world to the policy of Vietnam.' (McKay) 94. Namun ia sempat menyatakan prihatin atas concern on however 3SG still.be.able AS.evident.CAUS sikap anti-Asia itu attitude anti-Asia that 'However, he still has the opportunity to express concern about the anti-Asia attitude.' (McKay) In example 93, the emotion word simpati has a possessor (dunia), and it is also possible for an emotion word to be preceded by a quantifier, another typically nominal environment: 95. semua benci pada Caetano Calazar all hatred to Caetano Calazar ‘all the hatred of Caetano Calazar’ (McKay) 164 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Word bangga benci bimbang bosan cemas cemburu cinta curiga ganggu gemar gemas gila heran ingat jemu jengkel kangen kasih kasihan kenal kuatir lupa malu marah mimpi mirip peduli percaya Gloss proud hate anxious bored worried jealous love suspect upset fond of annoyed insane surprised remember fed up annoyed long for love pity know s.o. afraid forget ashamed angry dream resemble care about believe prihatin puas sadar salut sayang senang setia simpati suka takut tega yakin concerned satisfied aware respect pity, love like loyal sympathetic like afraid have the heart to sure Noun + + + ke- -an + + + + + + + + s.t. one is proud of i. hatred, ii. hostility i. doubt, ii. in doubt boredom anxiety jealousy i. love, ii. beloved suspicion + + + + i. fondness, ii. hobby i. anger, ii. passion madness miracle + + + + boredom annoyance what is missed pity + + + + anxiety forgetfulness shame anger + resemblance + + + + i. trust, ii. one who can be relied on concern satisfaction i. awareness, ii. realization + + + i. sorrow, ii. love, iii. darling i. happiness, favourite loyalty + + i. pleasure, ii. hobby fear + + + + + + + + + + + + + + + + + + + Gloss Table 4.3 – Indonesian Emotion and Cognition Words Use as nouns and nominalizations 165 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Many emotion and cognition words can also serve as the base for a nominal derived with the circumfix ke- -an: 96. Matanya lebar ketakutan eye.3 wide NOM.afraid.NOM ‘Her eyes were wide with fear.’ Table 4.3 shows which of the words listed in Table 4.1 are judged acceptable as nouns by informants on the basis that they can co-occur with the demonstrative itu, and which derive ke- -an nouns7. It is surprising that there is no obvious blocking effect where the root can be used as a noun. Several factors may play a role. Firstly, nominal uses of the roots are not common and they may be innovative. Several of the examples given above are taken from journalism and could be direct translations of English. Secondly, the morphologically derived nouns belong to a higher register than the zero derivations - there is a general tendency for morphological simplification in spoken Indonesian. Thirdly, some of the ke- -an derivatives do not have only the predictable meaning and this possibly allows the two forms to co-exist 8. 4.1.2.9 ATTRIBUTIVE USES Indonesian codes attributes in noun phrases in two ways: either as simple modifiers, or as predicative modifiers in relative clauses. In some cases, either possibility is grammatical: 97. rumah besar house big ‘the big house’ 98. rumah yang besar house REL big ‘the big house’ The structure in example 98 includes a relative clause, therefore any clausal predicate can occur in such structures and we would expect emotion and cognition words to be included. This is indeed the case: 99. orang-orang warga negara Turki yang simpati citizen state Turkey REL sympathetic person.DUP terhadap kelompok separatis Chechnya itu towards group separatist Chechnya that ‘the people who are sympathetic towards the Chechnya separatist movement’ (McKay) The circumfix ke- -an has other functions, as discussed in section 3.1.1; irrelevant derivations are ignored. 7 The secondary meanings of the ke- -an nominals often include a causative meaning although no causative morphology appears. Thus, kesenangan 'favourite' can be derived from senangkan 'make happy, make like'. This possibility appears elsewhere in the morphological system - see footnote 24 and section 3.3.1. 8 166 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 In general, verbs cannot be used as simple modifiers, but many emotion and cognition words can be used in such a structure. The most common collocation is with the root rasa used as a noun (compare the discussion of this word as a verb in the previous section): 100. Menurut Blainey, kaum penganggur ini membangkitkan class unemployed this AS.get.up.CAUS AS.follow Blainey rasa cemburu warga pembayar pajak negeri itu feeling jealousy citizen payer tax state that ‘According to Blainey, this class of unemployed gives rise to a feeling of jealousy in the tax-paying citizens of the country.’ (McKay) 101. Miss Nevill, yang telah mendapatkan kembali rasa return feeling Miss Nevill REL PAST AS.able.APPL percaya dirinya believe self.3 ‘Miss Nevill, who was able to recover a feeling of self-belief’ (SDPGS :113) Although these collocations are common, they should not be considered as compounds which provide the source of the verbal construction discussed in the previous section. If they were compounds, verbal suffixes would be expected to appear outside the second element, as is the case with the compound beritahu ‘inform, (LIT. give news)’. This compound root can take the causative suffix –kan outside the entire unit: 102. Ia sengaja memberitahukan kode kepada musuh 3SG deliberate AS.inform.CAUS code to enemy ‘He purposely revealed the code to the enemy.’ (E&S : 74) In contrast, the rasa + emotion or cognition word collocations do not behave in this way: 103. Saya merasakan 1SG AS.feel.APPL ‘I feel fear.’ 104. *Saya takut afraid merasatakutkan. Therefore there are two distinct ways in which the root rasa collocates with emotion and cognition words, as a verb with an attributive complement and as a noun with an attributive modifier. Emotion and cognition words do occur as modifiers of other head nouns, but these usages are less common: 167 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 105. Kedua, petani adalah pendukung setia Partai second farmer COP supporter loyal party Demokratik Liberal yang dikalahkan oleh democratic liberal by REL US.defeated.CAUS Hosokawa Hosokawa ‘Secondly, the farmers are loyal supporters of the Liberal Democratic Party which was defeated by Hosokawa.’ (McKay) 106. kemudian Barnes bertanya dengan nada then Barnes ask with tone ‘then Barnes asked in a surprised tone’ (SDPGS : 318) heran surprised Almost the same meaning as example 104 would be expressed if the word nada was omitted. The use of attributive words with the preposition dengan ‘with’ is very common in Indonesian. An example of this structure is the following: 107. Japp memandangnya dengan heran with surprised Japp AS.look.at.3 ‘Japp look at him, surprised.’ (SDPGS : 269) As will be discussed below, this construction is important in resolving the question of whether Indonesian has a lexical class of adjectives. The examples of this and the preceding section have shown that emotion and cognition words (or most of those considered here) can be used as nouns and in attributive constructions without morphological derivation. On the other hand the construction discussed in section 4.1.2.2 gives the appearance of using such words as verbs. Therefore the question of whether these words can usefully be assigned to a particular lexical category must be addressed, and it will be the topic of the next section. 4.1.3 Lexical categorisation Indonesian allows all major categories to act as the predicate of a clause. Many examples with verbal predicates have been seen in chapter 2, but in addition it is possible for nouns and prepositions to act as main predicate: 108. Pak Bodomo guru bagus Mr Bodomo teacher excellent 'Mr Bodomo is an excellent teacher.' 109. Ibu saya di Surabaya mother 1SG Surabaya LOC 'My mother is at Surabaya.' It is also possible for words which the English speaker thinks of as adjectives to act as predicates: 168 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 110. Rumah kita besar house 1PL.INCL big 'Our house is very big.' sekali very Therefore, it is not self-evident that emotion and cognition predicates should be assigned to the lexical category verb. Indeed, they do not consistently exhibit the morphological possibilities characteristic of verbs as discussed in chapter 2 (excepting those words which can occur as underived transitive verbs - see section 4.1.2.4), and in many respects their behaviour is reminiscent of that of English adjectives. For example, English constructions of the type I was angry with him are common, and control constructions are possible with emotion adjectives: I was happy to leave the city. The possibility that the Indonesian words should be treated as adjectives must therefore be considered. Before addressing this issue, however, two other questions must be resolved. Firstly, the same emotion and cognition words which function as predicates also appear in contexts which require nouns. As nouns can be the main predicate of a clause in Indonesian, the possibility that the words in question are basically nouns must also be considered. Section 4.1.3.1 considers and rejects this possibility. Secondly, the question must be asked: does Indonesian have the lexical category adjective at all? It is well known that languages without this category (or with a very restricted membership) exist (Dixon 1977) and a number of scholars (Cumming 1991, Thompson 1988) have argued that Indonesian is such a language, and that the words which express property concepts in Indonesian are a subclass of intransitive verbs. Section 4.1.3.2 examines the arguments and evidence relevant to this general question, and concludes that the distinction between adjective and verb does exist in Indonesian, although the two categories are more similar than they are in Indo-European languages. Section 4.1.3.3 then turns to discuss what the appropriate classification of emotion and cognition predicates might be. 4.1.3.1 NOMINAL USES OF EMOTION AND COGNITION WORDS There are three arguments against treating the construction in which an emotion or cognition word takes a bare DP complement as structures with nominal predicates. Two arguments are syntactic and one is semantic. The first syntactic argument hinges on the DP following the emotion word. If a true nominal predicate has a dependent nominal following it, this can almost always be treated as a possessor, and therefore an attached pronoun can be substituted for the dependent: 111. Pak Bodomo guru anak saya Mr Bodomo teacher child 1sg ‘Mr Bodomo is the teacher of my child’ 112. Pak Bodomo gurunya Mr Bodomo teacher.3 ‘Mr Bodomo is her teacher.’ 169 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 However, the bare DP complement of an emotion word can never be substituted by a reduced pronoun (examples 39-42 of chapter 2 repeated): 113. Saya 1SG sayang pity pada to 114. Saya 1SG sayang pity Siti Siti 115. *Saya 1SG 116. Saya menyayanginya 1SG AS.pity.APPL.3 'I pity Siti / her.' Siti Siti sayangnya pity.3 The impossibility of this type of structure was discussed in section 2.2.2.1 in relation to the properties of verbs; here it also provides evidence against an analysis of examples like 114 as involving nominal predicates. The second syntactic argument against this analysis is based on the type of negation used in clauses like example 114. As noted in section 4.1.2.7, the normal negator of nouns is bukan, and this is true for nominal predicates: 117. Pak Bodomo bukan guru Mr Bodomo NEG teacher ‘Mr Bodomo is not a teacher.’ 118. *Pak Bodomo tidak guru A clause such as example 114 would always normally be negated with tidak9: 119. Saya tidak sayang pity 1SG NEG ‘I don’t feel sorry for Siti.’ Siti Siti It is not possible to make a categorical statement here, as almost any Indonesian clause can be negated with bukan if particular emphasis on the negation is intended: 120. Saya bukan sayang Siti 1SG pity Siti NEG ‘I don’t feel sorry for Siti at all.’ However, in the unmarked case negation provides a reliable test which distinguishes nominal predications from other types of predicates, and this test shows that example 114 and similar clauses are not nominal predications. In more colloquial registers, other negators are such as nggak are also available. They have the same distribution as tidak and I ignore them in the following discussion. 9 170 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Finally, the semantic interpretation of nominal predications is not consistent with the meaning of clauses with emotion words and bare DP complements. Nominal predications in Indonesian are always equational, the two elements are understood as being identical: 121. Dokter itu wanita doctor that woman ‘The doctor is a woman.’ (MacDonald & Darjowidjojo 1967 : 213) 122. Hari ini hari gajian day this day pay ‘Today is pay day.’ (MacDonald & Darjowidjojo 1967 : 213) But such an interpretation is meaningless for a clause such as example 114: I am not identical to pity for Siti. The equational interpretation is equally impossible in the case of a clause with a cognition predicate: 123. kita lupa keyword yang sudah ditentukan 1.PL.INCL forget keyword REL PERF US.fixed.CAUS 'we forgot the Keyword which had just been established.' (McKay) We is not the same as forget the keyword ...; again the equational interpretation makes no sense. The arguments advanced in this section, from both syntax and semantics, show that although many emotion and cognition words can be used as nouns, as discussed in section 4.1.2.8, without morphological change, clauses in which they appear followed by a bare DP complement cannot be treated as nominal predications. 4.1.3.2 ADJECTIVES AND VERBS IN INDONESIAN Thompson (1988) gives explicit arguments that in Acehnese "property concepts do not form a separate lexical category from Verbs" (p169). She then includes Indonesian in a list of languages "for which similar arguments can be made" (p170), although this is immediately followed by the following qualification: That is, in each of these languages, whether or not a class of Adjectives can be identified on language-internal distributional grounds, Property Concept Words share many features with verbs. (p170) It is thus not clear exactly what claim Thompson is making with respect to Indonesian. I will initially assume that her claim is to be read in the strongest fashion, that is, that Indonesian is like Acehnese and has no separate lexical category adjective. Thompson offers four arguments for the lack of adjective as a lexical category in Acehnese, following Durie's analysis (1985: 101-103). The first argument is based on the possibilities for cross-referencing clitics and this is not applicable to Indonesian. The second argument is that Property Concept words can occur with aspect markers such as the inchoative ka. This 171 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 argument can also be made for Indonesian. Clauses in which a Property Concept word is the main predicate can have verbal auxiliaries10: 124. Risikonya akan berat sekali risk.3 heavy very FUT 'The risk will be very great.' (IRG: 235) 125. Minuman itu harus dingin drink that must cold 'The drinks must be cold.' (IRG: 235) Thompson's third argument is that verbs in Acehnese can be used freely to code attributes, that is there is no syntactic difference between the DP-internal modifiers in examples 126 and 127: 126. Acehnese aneuk muda nyan child young that 'that young child' 127. ureueng pula padè person plant rice 'that person planting rice' nyan that This argument is not true for Indonesian, at least not in the same way that it is true for Acehnese. Where a verb is used attributively within a noun phrase in Indonesian, it always follows the relativizer yang: 128. orang yang menanam person REL AS.plant 'that person planting rice' padi rice itu that Words attributing properties to a noun can appear in the same structure: 129. anak yang mudah young child REL 'that young child' itu that But in this case, the possibility also exists to place the Property Concept word in direct construction with the head noun: 130. anak mudah itu child young that 'that young child' But this is not possible with a verb; the equivalent structure is predicative11: Teeuw (1962) claims that Indonesian adjectives cannot co-occur with modals such as harus. His study is based on writings of a single author published between 1920 and 1950 (see also Teeuw 1977 for details of methodology). Examples 124 and 125 suggests that what is accepted as standard has changed over the intervening period. 10 172 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 131. ?orang menanam padi itu person rice that AS.plant 'The person is planting that rice.' Note that the structure here is radically different from that of example 128, where the demonstrative itu qualified the whole DP. In example 131, it only qualifies the object noun padi. Attributive constructions in Indonesian clearly differentiate between two classes of words, verbs and Property Concept words. Thompson's final argument from Acehnese is that many intransitive verbs which do not take an actor argument ('non-controlled verbs' in Durie's terminology) can be modified by the intensifier that 'very'. Indonesian does allow intensifiers with verbs (see example 68), therefore this argument is relevant: 132. Dia mengerti sekali very 3SG AS.undertsand 'She understood very well.' Clearly, the arguments which show that verbs and adjectives are not distinct lexical classes in Acehnese do not transfer convincingly to Indonesian. The claim that Thompson can hope to sustain is that made in the passage quoted above: that verbs and adjectives are distinct lexical classes, but are also very similar in some respects. I will now argue that this is the correct position12. Teeuw (1962) gives detailed morphological and syntactic criteria to justify a class of adjectives in Indonesian. The most important morphological criteria are the following: 1) addition of the prefix se- derives a comparative form. Thus besar 'big' gives sebesar as in sebesar telur ayam 'as big as a chicken's egg'. This prefix is impossible with verbal roots: beli 'buy' *> sebeli. 2) addition of the prefix ter- derives a superlative form. Thus terbesar 'biggest'. There are many verbs formed with a homophonous prefix. The meaning relation between the original verb and the derived form is that the degree of control of the actor is lessened in the derived form: bangun 'wake up' > terbangun 'be wakened suddenly, come to one's senses'. The importance of this prefix is also pointed out by Mahdi (1998). 3) addition of the 3rd person bound pronoun -nya derives a noun: The following example is marginal because the object is more definite than the subject. Such structures are avoided in Indonesian. 11 It should be noted that on the basis of the Durie/Thompson arguments for Acehnese the case of emotion and cognition predicates in that language is clearcut. Thompson's Property Concept words in Acehnese are non-controlled verbs, that is, they can only occur with optional undergoer enclitics and never with obligatory actor proclitics. But as seen in examples 11 and 12 above, there are emotion predicates in Acehnese which can take actor proclitics sometimes. They therefore do not form part of the same lexical class of verbs as Property Concept Words in Acehnese. 12 173 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 133. Tingginya orang itu seratus tujuh puluh high.3 person that one.hundred seven ten sentimeter centimetre 'That person's height is one hundred and seventy centimetres.' (IRG: 306) The bound pronoun -nya can only be attached to verbs with the prefixes meN- and di- (see extensive discussion in section 2.2.2.1), and it always codes a non-subject argument of the verb. A crucial syntactic criterion separating adjectives and verbs which Teeuw discusses is the possibility for adjectives to form adverbial phrases with the preposition dengan 'with': 134. Jumlah penduduk di kota Jakarta menanjak sum dweller city Jakarta AS.ascend LOC dengan cepat sekali with fast very 'Jakarta's population is rising very quickly.' (IRG: 208) This construction is not possible with verbs: 135. *Jumlah sum dengan with penduduk di dweller LOC mempercepat AS.per.fast kota city Jakarta Jakarta menanjak AS.ascend Nouns are possible following any preposition, including dengan, but other criteria establish a clear difference between adjectives and nouns. Firstly, the demonstratives ini and itu never modify an adjective or an adjective phrase. As adjectives follow the head in noun phrases, the two word classes are often adjacent, but without the head noun, or following a headless relative clause, their co-occurrence is ungrammatical: 136. rumah besar house big 'this big house' 137. yang 138. *besar big besar big REL 'this big one' ini this ini this ini this Similar restrictions apply to the co-occurrence of adjectives and numerals13. Teeuw (1962) thus establishes a distinction between verbs and the other two An additional criterion for Teeuw is that verbs can be preceded by modal auxiliaries, whilst adjectives cannot. See examples 124 and 125, and footnote 10 for contrary evidence from contemporary Indonesian. 13 174 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 major lexical classes, and a distinction between nouns and adjectives, thereby showing that adjectives are a lexical class in Indonesian. Where Thompson (1988) treats the similarities between verbs and property concept words by extending the class of verbs, Teeuw takes the opposite approach and argues that at least some verbs cannot be excluded from the class of adjectives. There is a class of intransitive verbs, traditionally referred to as basic verbs or grondwoordelijke werkwoorden in Dutch, which do not participate in the morphological alternations available to most verbs in the language. The class includes words such as datang ‘come’, turun ‘descend’ and tidur ‘sleep’. An early twentieth century source14 cited by Teeuw (1962: 416) even uses the term ‘adjectives of action’ for these words. The major consideration which influenced these scholars was that these verbs do not have forms with meN- or di-, nor do they occur in the Pro-V construction. As was argued in detail in chapter 2 of this work, however, these clause types are those available for verbs with two direct arguments. The morphological alternations involved signal the linking of actor or undergoer to the subject syntactic function, and this is not applicable to intransitive verbs except in a split-intransitive system. Such a system was discussed in relation to Acehnese data above, but clearly Indonesian does not have such a system. Therefore there is no reason to expect intransitive verbs to participate in the relevant morphological patterns. The other tests identified by Teeuw must therefore carry the full burden of telling us whether there are two word classes or only one, and for several tests, the evidence is unequivocal. There are no terprefixed derivatives of basic verbs having a superlative meaning, but some of them have such derivatives with a decontrolled meaning, for example tidur ‘sleep’ > tertidur ‘fall asleep’. Adverbial phrases with dengan are impossible for basic verbs unless there is a homophonous noun. So the phrase *dengan turun 'with descend' is not possible. Finally, Teeuw has a single example from his corpus of a basic verb deriving a form prefixed by se-: 139. Akan FUT tetapi but di LOC pasar market belum not.yet seturun se.descend itu that benar correct ‘But in the market it had not really gone down as much as that’ (Teeuw 1962 : 418) In general, such derivations do not seem to be possible. This evidence distinguishes the two classes clearly. Nevertheless, the point made by Thompson is once again shown to be valid: intransitive verbs and adjectives in Indonesian have a good deal in common. Having established that there is a lexical class of adjective in Indonesian, I now turn to the question of what category emotion and cognition words should be assigned to, whether it is useful to claim that they belong to a single category as lexical roots. Whatever the answer to this 14 C.A. Van Ophuisjen (1915) Maleische spraakkunst Leiden, publisher not known. 175 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 question, the further question can be asked whether this analysis can accommodate the clause type discussed in section 4.1.2.3 above. 4.1.3.3 THE LEXICAL CATEGORISATION OF EMOTION AND COGNITION WORDS The range of possible uses of emotion and cognition words discussed in section 4.1.2 immediately suggests that it will not be possible to analyse all of these uses as structures involving a single lexical category. In at least some cases, zero-derivation of different categories will be necessary. The best analysis will be the one which minimises the necessity for such derivations. Given the arguments advanced in section 4.1.3.1 and the fact that not all the words under consideration can be used felicitously as nouns, I will assume that treating the words as basically nouns is not a good analysis. The choice then is between analysing them as verbs or as adjectives. The criteria discussed in the previous section do not give an unequivocal answer, but they do favour the adjectival analysis for most of the words in question. As discussed in section 4.1.2.9, emotion and cognition words can be used attributively without the relativizer yang (examples 100, 101, 105 and 106) and they can form manner adverbials with the preposition dengan (example 107). The morphological behaviour of these words is also in general more adjectival than verbal, but the productivity of the adjectival morphology is limited. Of the words listed in Table 4.1, only the following have fully acceptable comparative forms with the prefix se-: 140. berani cemas gila marah mirip peduli prihatin setia takut tega brave worried mad angry similar concerned concerned loyal afraid heartless Speakers find clauses which use those derivatives which do exist to be very mannered, and would not use them under normal circumstances. Thus the following is grammatical: 141. Dia tidak setakut saya dengan 3SG NEG se.afraid 1SG with ‘(S)he is not as afraid of dogs as me.’ anjing dog but the meaning is more naturally expressed as follows: 142. Dia tidak takut dengan anjing 3SG NEG afraid with dog ‘(S)he is not as afraid of dogs as me.’ seperti as saya 1SG One word which fits the criteria for the group under discussion only appears as a predicate with the prefix se- attached. The root rupa means ‘form, kind’, and serupa means ‘to be similar’. 176 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 143. Ia kembali dengan sebuah buku yang serupa 3SG return with class book REL similar dengan kepunyaan Mr. Morley with possession Mr Morley ‘He returned with a book which resembled Mr Morley’s.’ (SDPGS : 50) I assume that this is a case of a homophonous prefix with a different function. The superlative with ter- also is only sporadically possible, although rather more productive than prefixation with se-. The following words have derivatives of this type: 144. bangga > terbannga berani > terberani cemas > tercemas cinta > tercinta gila > tergila jengkel > terjengkel kasih > terkasih khawatir > terkhawatir marah > termarah mirip > termirip peduli > terpeduli prihatin > terprihatin puas > terpuas sayang > tersayang senang > tersenang setia > tersetia takut > tertakut tega > tertega most proud most brave most worried most loved most crazy most disturbed most loved most worried most angry most similar most caring most concerned most satisfied most loved happiest most loyal most afraid most heartless Two points can be noted about these derivatives. Firstly, all the words which have se- derivatives also have ter- superlatives. These words therefore would seem to the prototypically adjectival members of the group. Secondly, some of the derivations are not regular. All three of the words which mean something like ‘love’, cinta, kasih and sayang, have derivatives in which the thematic role of the subject of the predication has switched, and indeed the glosses given in 144 are not quite accurate for these words. E&S give as a gloss for tercinta ‘beloved’, and it seems likely that the other two derivatives in this group are formed by analogy with this word and should have a similar translation. Two other words have similar derivatives: kenal ‘know’ > terkenal ‘well-known, famous’, and percaya ‘believe, trust’ > terpercaya ‘trustworthy’. In all these cases, the prefix ter- adds a meaning of intensity in these derivations, which is compatible with its general function with adjectives, but it also changes the relations encoded by the predicate. This leads to a further problem: as noted above the prefix ter- also derives verbs for which the participant’s control of the situation is decreased, to the extent that a passive is often the best English translation. A number of emotion and cognition words have derivatives of this type: 177 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 145. heran > terheran ingat > teringat lupa > terlupa pikir > terpikir sadar > tersadar astonished be reminded forget accidentally come to mind realize, recover consciousness An example of one of these derivatives is the following: 146. Untuk pertama kalinya Fandi teringat lagi for first time.3 Fandi ter.remember again pada Tuhan to Lord ‘For the first time, Fandi was reminded again of God.’ (SDM :272) An additional complication is that the syntactic valence of this derived form is apparently also not fixed. As well as taking a prepositional phrase complement, it can also appear with a bare DP: 147. Tiba-tiba saja dia teringat suaminya suddenly EMPH 3SG ter.remember husband.3 ‘All of a sudden, she was reminded of her husband.’ (SDM : 193) There are two possible explanations of this structure. Firstly, it may be an example of an oblique argument adjoined to the verb and licensed in that way rather than with a preposition. The other two cases in which I have argued that this occurs both involve a morphologically complex verb, which is a point of similarity with the current example. This analysis is discussed further in section 5.3.5 Secondly, it should be recalled that ingat is one of the words which can serve as a verbal root without an applicative suffix, and this fact may explain the last example15. One particularly interesting example is the word senang. This word occurs quite commonly in clauses with a single argument and is usually translated as ‘happy’ in such contexts: 148. Indonesia akan senang sekali bila PM datang happy very if PM arrive Indonesia FUT sebelum penyelenggaraan pertemuan itu before implementation meeting that 'Indonesia will be very happy if the PM arrives before the start of the meeting.' (McKay) The same word also occurs in clauses with a second argument coded as a prepositional phrase, and in such contexts it is usually translated as ‘like’: 15 teringat has other strange behaviour, described by IRG (116). 178 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 149. Aku senang rumah itu dan aku senang 1SG like house that and 1SG like ketenangannya peace.3 'I like the house and I like its peacefulness.' (E&S : 499) akan about Such translations suggest that there is a senang which is an adjective, and one which is a verb. But it is not entirely clear that the translation does not impose this analysis, rather than any facts of Indonesian. It is not hard to find examples which suggest strongly that the translation dichotomy is wrong: 150. Saya senang sekali atas hasil consolidasnya. like/happy very about result consolidation.3SG 1SG ?'I like very much the result of the merger.' 'I am very happy about the result of the merger.' (McKay) One can imagine a context where this would be translated as 'I like very much the success of the merger', but 'I am happy about the success of the merger' seems a good deal more natural. Senang also occurs with a complement clause, the subject position of which is controlled by the subject of senang. In many examples of such constructions, either translation is quite possible, with 'like' giving an habitual reading, and 'happy' suggesting a single occurrence: 151. Dia memang senang berdiam berlama-lama di she actually happy/like stay for a long time LOC bawah pancuran under shower 'She was actually happy to stay under the shower for a long while' OR 'She actually liked to stay under the shower for a long time' (SDM:8) Note that the difference in the translation here is basically one of aspect only, suggesting that the two concepts 'like' and 'happy' are close. There really seems no reason to assume two distinct words in Indonesian16. The evidence discussed in this section suggests that it may not be realistic to treat emotion and cognition words as though they all belonged to the same basic lexical category. A small group allow comparative and superlative formations like true adjectives, others can function as verbal roots without morphological derivation. The most realistic view is probably that there is a cline between adjectives and verbs in Indonesian, with some emotion and cognition words closer to one end of the continuum and some closer to the other. If this is the case, then the only useful approach would be to consider each word in detail and make a decision as to lexical category on a case by case basis, and this will not be attempted here. Rather, I turn to the question of whether the clause type in which an emotion and cognition word is followed by a bare noun phrase is a verbal clause or an adjectival clause. The case of suka makes it clear that these semantics are not arbitrary. Although it is not used as an intransitive predicate, suka is listed in the dictionary as a noun meaning 'joy, pleasure, happiness' as well as in its verbal sense, 'like'. 16 179 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 Previous studies (Kana 1986, Stevens 1970) took this for granted, as do the standard reference grammars (McDonald and Darjowidjojo 1967, IRG) in so far as they treat this type of clause at all. However, the discussion to this point indicates that the question is not straightforward, particularly as several of the most adjectival words, those with both comparative and superlative derivatives, can appear in this type of clause (gila, marah, mirip, takut). The crucial point is whether there is any evidence elsewhere in Indonesian that predicative adjectives can take a DP argument other than their subject. As far as I can discover, the only instance of this is in the comparative construction discussed previously (see example 141). In this case, the DP has a specific semantic role, object of comparison, and is presumably licensed by the morphology rather than by the adjectival base. There is evidence for this interpretation in the behaviour of the small group of words mentioned in this paragraph. For example, the word takut can take either a prepositional phrase or a noun phrase as complement: 152. Saya 1SG takut afraid dengan with 153. Saya takut anjing 1SG afraid dog ‘I am afraid of dogs.’ anjing dog When it takes the prefix se-, the object of comparison is a noun phrase, but the stimulus, the cause of the emotion, can only be a prepositional phrase: 154. Dia tidak setakut saya dengan se.afraid 1SG with 3SG NEG ‘(S)he is not as afraid of dogs as me.’ 155. *Dia tidak setakut saya anjing dog anjing This data reinforces the impression that the clause type with a prepositional phrase is adjectival and that with a noun phrase is verbal. An additional argument comes from the fact that the emotion and cognition words in the bare DP clause type share properties with a group of other words which are unequivocally verbal. Stevens (1970) discusses this evidence, and some of it will be important in the discussion of section 4.2. I have previously noted that a number of Stevens’ ‘pseudo-transitive’ verbs are used as auxiliaries (example 19 and related discussion), and one of the emotion and cognition words shares this property. suka ‘like’ can be used in this way, as can be seen from the possibility of it appearing to the left of the actor in a Pro-V clause: 156. Nasi itu suka saya makan eat rice that like 1SG ‘I will eat that rice.’ (Waruno Mahdi p.c.) That the verb is semantically bleached here is shown by the fact that the selectional restriction on its subject has been relaxed. When used as a verbal or adjectival predicate, suka requires a human subject, or an animate portrayed as capable of experiencing the feeling of liking. But in this use, it is 180 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 possible to have an inanimate subject. The other words with this property lack the possibility of appearing with a prepositional phrase complement (but see section 4.1.1 for discussion of ingin), but, with this exception, suka behaves exactly like them. Arguing by analogy, suka must be a verb at least in the clause type with a bare DP complement. One final point should be considered here. In a recent paper, Wierzbicka (1995) has argued that there are systematic differences in meaning between verbs and adjectives. She notes the fact that many emotion verbs in Russian have English adjectives as their closest translation equivalents, and claims that it is possible to detect a consistent difference in the degree of agency attributed to the experiencer across the two types of expression. A similar point is made in a study of adjective / verb minimal pairs by Pustet (2000). If such observations are true, it would be surprising if the prepositional complement clause type and the bare DP complement clause type should be headed by respectively an adjective and a verb and still have apparently identical meaning (see section 4.1.2.3 for a relevant example). This might be taken as an argument for treating the prepositional complement clause type as verbal also, although I would be reluctant to make this move for reasons which will be discussed in chapter 5. The argument from semantics is less compelling if we note that the studies referred to above deal with translation equivalents from two languages and pairs of words within a single language. But the data under discussion consists of words which may appear in two lexical categories without any morphological derivation. In such cases, it is legitimate to ask how much the semantic structure of the two items might differ. The null hypothesis in this case must be that where the meaning of a word is compatible with the semantic constraints associated with constructions into which two lexical categories can be inserted, then the word can be inserted into either with no change, or minimal change, to its semantic structure. The case of senang, discussed above, is very relevant. Although outside the scope of this study, it should not be too difficult to provide a semantics for this word which is compatible with both the intransitive uses and the transitive ones. That the result will not correspond closely to either like or happy in English is irrelevant. To summarise this discussion, there is evidence that many, even most, emotion and cognition words have adjectival characteristics in Indonesian and the clause type which I take as criterial for membership of the word class is an adjectival construction. But many of the words also appear in the bare DP clause type, and the evidence just discussed suggests strongly that this is a verbal construction. The next section will take this as an assumption in its discussion, and chapter 5 will present arguments that the properties of that clause type only make sense within the full picture of Indonesian clausal syntax if that assumption is maintained. 4.2 Emotion verbs and grammatical functions This section deals with the group of Indonesian verbs identified in the previous section and examines their status in regard to the syntactic transitivity system of that language. These verbs do not appear in the range of construction types analysed in chapter 2 and which I have assumed to be 181 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 typical of transitive verbs in Indonesian. The question this section seeks to answer is what grammatical function is assigned to the stimulus argument in a clause headed by an emotion or cognition verb and including two bare DPs (see examples in section 4.1.2.3 above). I attempt to answer this question using the theoretical resources of Lexical Functional Grammar (LFG). This framework assumes a constrained list of grammatical functions (GFs): subject (SUBJ), object (OBJ), oblique (OBLΤ) and second object (OBJΤ). The last two GFs are subscripted with a theta because in the case of OBJΤs their thematic role is assumed to be restricted within any particular language (Bresnan and Kanerva 1989), and in the case of OBLΤs their thematic role is indexed by the adposition or case marker which licenses them. These GFs are not configurationally defined. They are elements of f(unctional)-structure, a structure which is distinct from but parallel to c(onstituent)-structure (see section 1.3.1 and Bresnan 2001a: 50-56). Syntactic properties other than configurational ones are therefore of primary interest in this investigation. The group of predicates of interest here includes the following words: 157. berang bosan gila kangen kasih kasihan kuatir lupa angry, irate bored crazy, obsessed miss, long for love love, pity fear forget marah angry mimpi mirip peduli sayang senang suka takut yakin dream resemble care about, pay attention love, pity happy, like like fear sure This list is not exhaustive; rather it includes just those words for which I have clear examples from texts. Also, several words are left out of this list which might have a claim to be included. These words (benci 'hate', ingat 'remember' and percaya 'believe') all occur in the construction under consideration. But they also all have related transitive verbs which are not derived with an applicative suffix (membenci, mengingat and mempercaya). Therefore it is not possible to say whether clauses with these predicates and a DP stimulus are examples of the relevant construction, or whether they are rather examples of the unprefixed transitive verb construction discussed in chapter 2. Given this unclarity, it is preferable to omit these predicates from the discussion. Section 4.2.1 examines the evidence that the emotion and cognition verb clauses are distinct from normal transitive clauses in Indonesian, and shows that the stimulus is neither a SUBJ nor an OBJ. Sections 4.2.2 and 4.2.3 examine evidence which might show that the stimulus is an OBLΤ or an OBJΤ respectively, concluding that the second possibility is to be preferred. 4.2.1 Transitive Clauses - SUBJ and OBJ The overwhelming majority of transitive verbs in Indonesian can appear in four types of clause: 182 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 158. Dia 3SG membaca AS.read buku book itu that 'S/he read the book.' 159. Buku book itu that dibaca US.read (oleh) (by) Ali Ali 'The book was read by Ali.' 160. Buku book itu saya that 1SG baca read 'The book, I read.' 161. Dia 3SG baca read buku book itu that 'S/he read the book.' Various restrictions apply to the constructions exemplified in 158 to 161; these have been discussed in detail in chapter 2. Clauses with emotion predicates look the same as the type exemplified in 161, but the other clause types are not possible: 162. Aku bosan hidup terus-menerus di Surabaya 1SG bored life continue-AS.continue Surabaya LOC 'I am fed up with living all the time in Surabaya.' (PYD: 180) 163. *Aku membosan hidup terus-menerus di Surabaya. The type of clause illustrated in example 159 is not possible with a first person actor, but substitution of a third person actor does not make this construction possible with this predicate: 164. *Hidup terus-menerus di Surabaya dibosannya. The type exemplified in 160 is not possible either, although the direct manipulation of example 162 yields a sentence which does have a reading: 165. Hidup terus-menerus di Surabaya aku bosan. ‘Living all the time in Surabaya, I am fed up with it.' This clause requires an intonation distinct from that associated with clauses of the type seen in example 160 with a pause after the initial DP which is a topicalized constituent, and the addition of an auxiliary verb to the sentence makes the distinction absolutely clear. In a construction of the type in example 160, the actor pronoun remains adjacent to the main verb even when an auxiliary or modal is included, but in emotion predicate clauses, the actor pronoun appears to the left of an auxiliary: 166. Buku itu akan book that FUT 'I will read that book.' saya 1SG baca. read 183 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 167. Hidup terus-menerus di Surabaya aku akan bosan. 'This monotonous life in Surabaya, I will be fed up with it.' 168. *Hidup terus-menerus di Surabaya akan aku bosan. This evidence shows that these predicates do not participate in the system of clauses associated with verbs assumed to be transitive in Indonesian. Therefore it is not possible to describe the emotion and cognition verbs as syntactically transitive. This status alone may be enough to show that the stimulus argument of an emotion predicate is not an OBJ. If OBJ is defined as the non- SUBJ argument of a transitive verb, then stimuli are not objects because the predicate that they are associated with is not syntactically transitive17. A more stringent test requires OBJs be able to become the SUBJ of a related clause, typically a passive. Chapter 2 has provided detailed arguments as to whether the relationships between the possible clause types for transitive verbs in Indonesian can be compared to the active-passive relation in a language like English. Leaving aside that issue, it is clear that stimuli also fail this test as seen in example 164. Many of the emotion predicates derive true transitive verbs with an applicative suffix, and a comparison of their properties with the basic predicates is instructive. The applicativized verbs appear with both meN- and di- prefixes: 169. Dia merasa perempuan itu sedang mengasihaninya 3SG AS.feel woman that PROG AS.pity.APPL.3 'She feels that that woman is pitying her.' (SDM: 86) 170. Dan dia tidak suka dikasihani and 3SG NEG like US.pity.APPL 'And she does not like to be pitied.' (SDM: 86) Further examples are given above in section 4.1.2.4. The contrast in syntactic behaviour between the derived transitive verb which does have an OBJ and the emotion predicate is clear. The possibility that the stimulus is the SUBJ of its clause can also be eliminated for various reasons. Firstly, it would be surprising that an argument alternated between being an OBLΤ (in the prepositional construction) and a SUBJ with no verbal morphology. Secondly, it can be seen from example 169 that the experiencer is the SUBJ of the corresponding applicative verb with meN- prefix, the clause type which corresponds to an active clause. Applicativization does not affect SUBJs, therefore the experiencer should be the SUBJ of the basic predicate. A clause can only have one SUBJ, therefore the stimulus cannot be one. Further, Indonesian has a third person pronoun, ia, which is restricted (more or less) to appear in SUBJ position. This pronoun can occur as experiencer of an emotion predicate clause: I assume that, as SUBJ and OBJ are syntactic concepts, syntactic transitivity is the relevant notion for this argument. 17 184 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 171. Ia kuatir suratnya tidak sampai 3SG fear letter.3 arrive NEG 'She was afraid her letter would not arrive.' (ES: 313) Once again, the fact that a clause can only have one SUBJ forces the conclusion that the stimulus cannot be SUBJ. Finally, the examples in section 4.1.2.7 with complement clauses with a controlled subject position showed that the experiencer was the sole nominal argument of the emotion predicate, and controller of the reference of the missing argument. It must therefore be SUBJ in such examples. This section has demonstrated that emotion predicates are not transitive verbs in Indonesian. Therefore, a bare DP stimulus argument following such a predicate cannot be an OBJ. And the (remote) possibility that such arguments are SUBJs has also been eliminated. 4.2.2 OBLΤ Having established that a DP stimulus of an emotion predicate cannot be assigned the GFs SUBJ or OBJ, I now turn to examine the arguments as to whether such elements might be classified as OBLΤ. I have discussed above (section 4.1.2.3) the apparent lack of semantic differentiation between the clause type with a prepositional complement and that with a bare DP complement. Another example (see also examples 58 and 59) which makes this clear is the following, in which the two types are co-ordinated: 172. Aku senang rumah ini, dan aku senang like house this and 1SG like 1SG akan ketenangannya about NOM.quiet.NOM.3 'I like this house, and I like it's peacefulness.' (E&S: 499) Such a difference in syntactic coding is normally correlated with a semantic difference. Therefore, it might be possible to argue from the lack of semantic differentiation that the stimulus clauses with a bare DP complement is still an OBLΤ, but one with non-standard coding. There are two pieces of evidence which suggest that this would be a wrong conclusion. But before presenting this evidence, there is once more a definitional issue. Indonesian is not a case-marking language, and therefore the obvious distinction between core and non-core arguments is the presence of prepositions. This is not a definitive criterion however. The literature includes examples both of DPs with the coding properties of obliques being direct arguments, and of DPs with the coding properties of direct arguments being obliques. Examples of the first type are animate, referential objects in Spanish, which are preceded by the preposition a: 173. Busco a mi amigo seek.1SG a 1SG.POSS friend 'I'm looking for my friend.' (Hopper & Thompson 1980: ex11b) and various subjects in Icelandic with 'quirky' case-marking: 185 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 174. leiδ inlegur boring.NOM.SG Henni hefur alltaf þótt Ólafur she.DAT has always thought Olaf.NOM 'She has always considered Olaf boring.' (Zaenen, Maling & Thráinsson 1985: ex13)) An example of the second type is the use of accusative case in Icelandic to mark some temporal adjuncts (Smith 1996: 40-42): 175. Ég var þar tvo daga 1SG.NOM was there two.ACC days.ACC 'I was there for two days.' (Smith 1996: 41, ex49) Also relevant here is the syncretism of ergative case, which I assume marks terms, with some oblique case, typically instrumental or genitive, in many languages with ergative-absolutive morphology (Blake 1977: 60). The oblique arguments adjoined to verbs discussed in chapters 2 and 3 are also examples of this type. Such examples show that coding properties alone are not sufficient to establish the syntactic status of an argument. Behavioural tests are necessary also, and I now turn to two such tests which show that the stimulus of the first conjunct of example 172 and similar clauses does not have the properties of an oblique argument. The first way in which DP stimuli behave like direct arguments is that a quantifier can be floated from them. This is not possible with a PP stimulus, an oblique argument, or a PP which is an adjunct: 176. Anak-anak itu suka gula-gula that like sugar.DUP child.DUP 'All the children like the sweets.' OR 'The children like all the sweets.' itu that semuanya. all 177. Anak-anak itu suka dengan that like with child.DUP 'All the children like the sweets.' NOT 'The children like all the sweets.' 178. Orang-orang Sasak datang dengan anak-anaknya man.DUP Sasak come with child.DUP.3 semuanya all 'All the Sasak people came with their children.' NOT *'The Sasak people came with all their children.' gula-gula sugar.DUP itu that semuanya all In each case, the quantifier can be read as having floated from the SUBJ, and evidence to be presented below will show that other direct arguments also have this property. But only with the DP complement of an emotion verb can the quantifier be read as having floated from the other DP in the clause; when that DP is within a PP the quantifier cannot be construed with it18. Section One of my consultants allows both readings of example 177, but the second reading seems to be marginal and is strongly rejected by the other consultants. 18 186 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 2.3.1.3 has presented evidence that the adjoined oblique actor in a di-V-DP clause cannot float a quantifier, as has section 3.2.2.1 for the non-subject argument of ke- -an verbs . The second piece of evidence comes from extraction facts. Indonesian subjects can always be extracted leaving a gap and this generalisation is agreed on by all sources. The facts regarding other arguments of transitive verbs are in dispute; they have already been mentioned in section 2.2.2.2 and will be discussed in detail in chapter 5. For emotion predicates, I am only aware of one discussion of extraction in the literature, Stevens (1970), and this source claims that extraction of DP stimuli is grammatical19. This judgment is shared by the native speakers I have consulted. Gapped extraction from PPs is completely impossible however20, neither prepositionstranding nor pied-piping is available in Indonesian. The status of the PP involved, argument or oblique, is not relevant: 179. gula-gula yang anak-anak child.DUP sugar.DUP REL 'the sweets that the children like' suka like itu that 180. *gula-gula sugar.DUP REL anak-anak child.DUP suka like dengan with *gula-gula sugar.DUP dengan with yang anak-anak child.DUP 181. 182. yang REL suka like itu that itu that *anak-anak yang orang Sasak datang dengan itu FOR: 'the children that the Sasak people came with' Section 2.2.2.2 has presented evidence that the adjoined oblique in a di-V-DP clause cannot be relativized, and section 3.2.2.1 has made the same point in respect of the non-subject argument of adversative verbs. Clearly, the behavioural properties of DP stimuli are different from those associated with the grammatical function OBLΤ in Indonesian, regardless of whether the OBLΤ is coded as a PP or adjoined to the verb. Therefore it is not possible to maintain an analysis which treats these constituents as obliques, and it seems that the only possible analysis within LFG's inventory of grammatical functions is to treat the stimuli as OBJΤs. I turn to an examination of the evidence for this position in the following section. The data presented in section 4.1.2.2 suggesting that in the prepositional construction the stimulus is assigned a semantic role directly by the predicate, with the preposition making no semantic contribution may also be relevant here. This behaviour is not typical of true OBLΤs, which display semantic interdependence with the preposition or case-marker which licenses them. Therefore this evidence suggests that rather than the bare DPs being Kana (1986) discusses emotion predicates in some detail (see also below), but does not consider extraction. 19 20 Extraction with a resumptive pronoun is possible: see IRG :289-291 and section 5.1.1. 187 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 non-standard OBLΤs, it is actually the arguments coded as OBLΤs which are nonstandard. The head of the DP is the semantic head of the whole constituent. For this reason, when developing an explicit analysis of the emotion and cognition verbs in section 4.3, I treat these prepositions as case-markers rather than as heads. 4.2.3 OBJΤ As mentioned above, OBJΤ is the grammatical function of the second object of ditransitive verbs for LFG. Indonesian has a small number of underived verbs which are ditransitive, and a large number derived with the applicative suffixes -i and -kan. I will discuss here three properties of the second objects of such verbs: access to SUBJ in a related clause, quantifier float and extraction and compare them with the properties of DP stimuli. The second object of a ditransitive clause with a meN- prefixed verb cannot become subject of a clause with a di- prefixed verb: 183. Ali mengirimi wanita itu Ali AS.send.APPL wanita that 'Ali sent the woman the letter.' surat letter itu that 184. *Surat itu dikirimi wanita itu letter that US.send.APPL woman that ('The letter was sent the woman.) (Chung 1976a: ex.63a) In the case of derived ditransitive verbs such as that in example 182, the same thematic argument would be OBJ in a related clause with an underived, meNprefixed verb and has access to SUBJ function in a clause with a di-prefixed unsuffixed verb: 185. Surat itu dikirim kepada wanita woman letter that US.send to 'The letter was sent to the woman.' itu that The motivation for applicative is to make the third argument available to syntactic processes which require core status; there is no motivation to maintain the status of other arguments. In respect of access to SUBJ, the behaviour of the OBJΤs and the DP stimuli is similar. DP stimuli cannot become SUBJ of a related clause either unless the verb is morphologically altered. They can become SUBJ of a related clause headed by a di-prefixed derived verb (example 170) and the behaviour of the two argument types diverges here. Second objects cannot be SUBJ or OBJ with derived verbs while stimuli can. However, the relevant verbal derivation, applicative, promotes an oblique argument to core status, therefore the source of the applicativized emotion verbs must be the construction with a PP stimulus and the comparison between the two types is not in fact direct on this point 21. I ignore here the issue of applicative derivations in Indonesian which take a transitive verb and derive another transitive verb with a different meaning e.g. memergok 'catch someone by surprise', memergoki 'catch someone redhanded'. 21 188 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 In the previous section, examples were given which showed that a quantifier could float from a SUBJ. This is also possible from OBJs and from OBJΤs: 186. Saya mukul anak-anak itu 1SG hit child.DUP that 'I hit all the children yesterday.' 187. Saya memberinya hadiah 1SG gift AS.give 'I gave her all the presents.' kemarin yesterday itu that semuanya all semuanya all The descriptive generalisation is that quantifiers can float from any term argument in Indonesian. And as previously shown, a quantifier can also float from an DP stimulus (example 176, repeated here): 188. Anak-anak itu suka gula-gula child.DUP that like sugar.DUP 'All the children like the sweets.' OR 'The children like all the sweets.' itu that semuanya. all In this case, the stimuli behave like term arguments; they are not so OBJΤ is the remaining possibility. SUBJ or OBJ, Many descriptions of Indonesian state that extraction (relative clause formation and question formation) is only possible with SUBJs. The facts are a good deal more complicated than this as soon as anything other than the prescriptive standard is investigated; a more complete account is given by Voskuil (1996: Ch 8), see also sections 2.2.2.2 and 5.1.The discussion of chapter 2 suggests that the correct generalisation is that SUBJ and OBJ can be extracted in the absence of a verb prefix. Thus an OBJ can be extracted from a clause of the type seen in example 161 (auxiliary added to disambiguate the construction)22: 189. buku yang dia akan book REL 3SG FUT 'the book that she will read' baca read itu that Chapter 2 did not consider the extractability of second objects, and judgments vary for these arguments. As might be expected, such arguments cannot be extracted when the verb carries a prefix, and no native speaker I have consulted will allow extraction when the verb is derived with an applicative suffix. In the case of clauses with underived, unprefixed ditransitive verbs, however, at least some speakers will permit extraction of the second object23: Michael Ewing (p.c.) informs me that he has observed numerous examples of such constructions in recorded conversation of educated Indonesians. 22 This is not true for the verb ajar 'teach' for my primary consultant. I have no explanation of this fact. 23 189 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 190. Saya akan beri dia 1SG give 3SG FUT 'I will give him the book.' 191. ?buku yang saya akan book REL 1SG FUT 'the book that I will give him' buku book beri give itu that dia 3SG itu that As demonstrated above (example 179), this type of extraction is also possible with a DP stimulus: 192. orang yang saya person REL 1SG 'the person that I like' suka like itu that On this test, the DP stimulus possesses the syntactic properties of a direct argument more clearly than OBJΤs. I have demonstrated that DP stimuli cannot be plausibly assigned any of the other three GFs recognised by LFG, and that their syntactic properties are very similar to those of OBJΤs. Within the theory as currently formulated, this would seem to be the only possible analysis. Section 3.3.2.3 has already raised the possibility of LMT predicting such an array of GFs, although in that case the data showed that it could not be correct. But in the case of emotion and cognition verbs, the data do support such an analysis. The only alternative is to add a GF to the inventory, and the final section of this chapter will examine which of these two possibilities is preferable from a theoretical point of view. First, I discuss the formal issues which arise in implementing such an analysis. 4.3 Formalising the analysis 4.3.1 An LFG analysis There are two issues to be addressed in analysing the Indonesian emotion and cognition verbs using the apparatus of Lexical Mapping Theory. The obvious question is what feature should be lexically specified for the second argument in order to ensure that it is mapped to the OBJθ GF. There are two possibilities, and these will be discussed in section 4.3.1.2. Before that, I deal with a less obvious question: what is the status of the first argument, the experiencer, of these verbs? 4.3.1.1 TRANSITIVE OR TWO-PLACE UNACCUSATIVE The principles in LMT which assign features to semantic roles give particular emphasis to patientlike roles (Bresnan 2001a: 309). The first such role in the a-structure is assigned the feature [-r], with other similar roles assigned the value [+o]24. These principles have two effects. Firstly, they ensure that the arguments of ditransitive verbs are linked correctly. The first object will have the feature [-r] and the lowest role it can link to is thus OBJ. I follow the principles given by Bresnan (2001a: 309) for English. As far as I can tell, these principles also work for Indonesian. 24 190 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 The second object, in a language such as English or Indonesian which does not allow symmetrical objects (Bresnan and Moshi 1990), will have the feature [+o], and the lowest role with which it is compatible is OBJθ. The second effect is to make a distinction between unaccusative and unergative intransitive verbs. Because unaccusatives have a patientlike argument, they are assigned the feature [-r], while the argument of unergatives receive the feature [-o] by a default principle. While this distinction has no significance in the mapping to GFs (the argument of all intransitive verbs maps to SUBJ), it does allow other effects to be explained neatly (see Bresnan 2001a: 312-314 for discussion of unaccusatives and resultatives). This analysis of intransitive verbs allows the possibility that two-place unaccusative verbs might exist. This possibility has been discussed in detail within the P&P framework in Belletti and Rizzi (1988) in an analysis of Italian psych verbs to which I will return. Note that in the current framework, such verbs would have exactly the properties I am suggesting the Indonesian emotion and cognition verbs possess. The first argument in the a-structure, being patientlike will be assigned the feature [-r]. The second argument is also patientlike, and given that the language does not allow symmetric objects, the only possible feature assignment is [+o]. In the absence of a logical subject with the feature [-o], a [-r] argument can map to SUBJ and therefore the first argument will do so. The second argument then maps to the lowest role compatible with its feature specification, which is OBJθ. The attraction of this analysis is that once the type of the verb is specified, everything else falls out from general principles of the theory. If this analysis were correct, we would predict that the emotion and cognition predicates which can have a bare DP second argument would all be unaccusatives. It is then a question of some importance for this discussion whether some or all emotion and cognition verbs in Indonesian can be considered as unaccusatives. The suggestion is not preposterous on semantic grounds. The existence of verbs in many languages which code the experiencer as object, such as frighten in English, shows that it is possible to construe an experiencer as affected by the event denoted by the verb. Some accounts emphasise that in such cases there are reasons from the semantics of the arguments (Pesetsky 1995) or the aspectual structure of the event (Grimshaw 1990) to take the experiencer as the less prominent argument. But Dowty (1991) discusses the respective proto-Agent and proto-Patient entailments of the two arguments of verbs denoting psychological processes, and concludes that both have something in common with each proto-role. Being affected by the event is the entailment which Dowty sees as most clearly identifying the experiencer with the proto-Patient role. In addition, the aforementioned study by Belletti and Rizzi (1988) sees no inconsistency in treating at least some psych verbs as having two internal arguments. Kana (1986: chapter 3) identifies two morphological tests which she claims separate unergative and unaccusative verbs in Indonesian. Firstly, she notes that the suffix -kan can be used to add a second argument to some intransitive verbs. In some cases, the additional argument is an object: 191 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 193. Pegawai itu bekerja di kantor employee that ber.work LOC office 'That employee works in the governor's office.' gubernur governor 194. Sri mengerjakan pekerjaannya dengan baik Sri AS.work.APPL work.3 with good 'Sri carried out her work well.' (Kana 1986: 56, ex20b,a) In other cases, it is a subject: 195. Dia tidur di atas tikar straw.mat 3 sleep LOC top 'She sleeps on a straw mat.' 196. Ibu menidurkan anaknya child.3 mother AS.sleep.CAUS 'Mother put her child to sleep.' (Kana 1986: 57, ex.26b,a) As my glosses above indicate, the traditional analysis is that the suffix has two functions, as an applicative and as a causative. Kana claims that the suffix is a transitivizer whose effect depends on the valence of the base verb. Where the base is unergative, the suffix adds an object (i.e. applicative), and where the base is unaccusative it adds a subject (i.e. causative)25. Similar correlations have been noted for multi-functional affixes in other languages (Austin 1997a). Secondly, Kana claims that out of a range of nominalising morphologies available in Indonesian one is restricted to being used with verbs which have a patient (an initial 2 for Kana) in their lexical structure. The circumfix26 peN- -an derives nouns from verbs which have the meaning of 'the action of Ving', and they can take the undergoer of the original verb as an argument of the noun: 197. 198. bunuh kill sampai arrive pembunuhan serdadu soldier 'murder of the soldier' NOM.kill.NOM penyampaian surat letter 'delivery of a letter' NOM.arrive.NOM Kana does not provide detailed negative evidence establishing that the verbs she identifies with this test are unaccusative, but when such evidence is examined it suggests that this test is not reliable, or that large numbers of Indonesian verbs have multiple lexical entries. Several of the verbs which Kana uses to exemplify the function of -kan with unergatives turn out to have peN- an nominals derived from them: 25 For a different account, which also includes the suffix -i, see Voskuil 1996. The capital N here has the same meaning as it does in the actor subject prefix meN-: the nasal undergoes various assimilations to following consonants. See chapter 1 for details. 26 192 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 199. Orang tua saya akan bercerai person old 1SG ber.separate FUT 'My parents are going to get divorced.' 200. Anwar menceraikan isterinya Anwar AS.separate.APPL wife.3 'Anwar divorced his wife.' (Kana 1986: 60, ex34b,a) 201. penceraian - divorce, separation 202. Saya ingin berbicara 1SG wish ber.speak 'I want to speak with you.' 203. Mereka membicarakan soal itu matter that 3PL AS.speak.APPL 'They discussed the problem.' (Kana 1986: 60, ex37b,a) 204. pembicaraan - discussion dengan with saudara cousin Given such data, and the fact that peN- -an nominalisation is not widespread for emotion and cognition verbs, I will not use this test in what follows27. This leaves the evidence of -kan suffixation as the only test for the status of these verbs. For the purposes of the following discussion, it is necessary to assume either that uses of emotion and cognition verbs with a single argument are verbal rather than adjectival contrary to the conclusions of section 4.1.4.2, or that adjectival predicates in Indonesian take a subject and that this argument has a semantic role which is not agentlike, that is, that they are unaccusative predicates. As discussed in section 4.1.2.4, some of the Indonesian words under consideration do form causatives with -kan. Very often the use of such derived verbs is quasi-adjectival, with only a single argument: 205. Orang Inggris adalah orang yang person England be person REL membosankan AS.boring.CAUS 'English people are very boring.' (SDPGS: 92) sangat very But examples do occur in which they do take two arguments and it is clear that the object of the causative verb is the experiencer: Two of the verbs which are potential unaccusatives, kenal and yakin do have peN- -an nominalizations. Interestingly, in each case the nominal would seem to be based on the causative verb derived from the base even though the -kan suffix does not appear in the nominal. Thus, kenal 'know' has a causative kenalkan 'introduce' and pengenalan means 'introduction, and yakin 'sure' has a causative yakinkan 'convince' and peyakinan means 'act of convincing'. If this possibility is systematic, Kana's claim as to the nominalisation possibilities of unaccusatives might be dependent on the parallel possibility of causativization. See also footnote 6 and section 3.3.1 for the morphological issue. 27 193 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 206. Ia menyukakan anaknya pada anjing 3SG AS.like.CAUS child.3 to dog 'She taught her child to like dogs.' (E&S: 530) 207. Bermimpi jahat tentu menakutkannya dream evil certainly AS.afraid.CAUS.3 'He surely was frightened by a nightmare.' (E&S:544) As discussed in section 4.1.2.4, there are also emotion and cognition verbs which use the suffix -kan with applicative effect. However, closer examination suggests that there may be a degree of correlation between being able to take a bare DP second argument and forming causative derived verbs with -kan. There are five verbs which are apparent counter-examples to this correlation. One of them, ingat 'remember', has a corresponding unsuffixed transitive verb. It is therefore possible that the causative derivation is based on this verb. Another potential counter-example is the verb takut 'fear', and as discussed in section 4.1.2.4, this verb can form applicatives and causatives with both suffixes, -i and -kan. It would be hard to base any useful argument on the behaviour of this verb. Of the remaining three verbs, for one, peduli 'care, attend', I have a single example in which it takes a bare DP second argument which is also not reliable evidence. The remaining two verbs, lupa 'forget' and mimpi 'dream' both provide more serious counter-evidence. lupa appears with a bare DP following in several examples: 208. Kamu lupa rumahku? forget house.1SG 2SG 'Have you forgotten my house?' (SDM: 96) and it is quite common with the -kan suffix (21 instances in my database), always with the experiencer as subject. The verb mimpi is less common, but the facts seem to be similar. These two verbs then are real problems for the unaccusative analysis. One possibility is to take lupa and mimpi (and others which I have not come across, no doubt) to be a separate class of verbs for which the experiencer is treated as agentlike. To do this would in turn raise the question of whether there was any semantic basis for this classification, or whether it was a case of lexical idiosyncrasy. The division between unergative and unaccusative verbs is notoriously idiosyncratic (Rosen 1984), and this might be a possible account. It would still be necessary, however, to also give an account of the group of verbs which take a bare DP second argument and form applicatives with -i. Kana (1986) does not claim that this suffix can only attach to intransitives of one type. Her claim is rather that -i signals that a locative has been advanced to object status. This is an additional complication for the two-place unaccusative account. Three classes of verbs must be recognised: transitives with a nonlocative second argument (at least lupa and mimpi), two-place unaccusatives (the verbs with -kan causatives) and verbs with a locative as second argument (those with -i applicatives). The status of the first argument of this last group 194 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 of verbs would be unresolved; either they are not unaccusative because they do not take -kan as an applicative suffix, or locative advancement takes priority in the morphological rules. There is however good evidence to reject the analysis of the stimulus argument as an underlying locative. As documented in detail in section 4.1.2.2, the true locative preposition di never occurs with emotion and cognition predicates, and the other prepositions which code static location in Indonesian are also almost unrepresented. Together with the fact that for almost every one of the predicates listed in Table 4.2 both pada and dengan are possible, this means that an extremely liberal interpretation of the notion locative would be needed to carry the analysis through. An alternative, and to me, preferable analysis is to assume that the emotion and cognition verbs form a unified class, and that they are genuinely transitive, that is, that they have an external argument. This means accepting that there is some idiosyncrasy in the uses of the suffix -kan, an unremarkable proposition when the behaviour of the verb takut is recalled, and that the correlation between emotion and cognition verbs and -kan applicatives is suggestive but not very significant. 4.3.1.2 LEXICAL FEATURES OF THE STIMULUS Following the arguments of the previous section, I will treat Indonesian emotion and cognition verbs as two-place verbs with an external argument. This means that the first argument in the a-structure will not be taken as patientlike and associated with the value [-r]. But the second argument cannot be associated with that value either, or it would map to the OBJ GF, which is not the desired result. In order for the second argument to map to the OBJθ function, it must be associated with either the value [+r] or the value [+o]. The argument must map to the most marked compatible function: if the value [+r] is associated with it the options are OBJθ and OBLΤ, if the value [+o] is associated the options are OBJ or OBJθ. In either case, OBJθ is more marked and will be the function mapped to. The question is whether there is any reason to prefer specifying one feature in the lexicon rather than the other. I will argue here that there is a strong reason for preferring to specify the feature [+r], and this argument is based on one of the claims of LMT, that there are natural classes of argument functions. In section 4.1.2.2, I argued that even when emotion and cognition predicates are followed by a prepositional phrase, they assign a semantic role to the argument inside the PP. The essential difference between that construction and the construction with a bare DP complement lies in the ability or inability of the predicate to syntactically license a second argument. Depending on one's view of the evidence discussed in section 4.2, this is dependent either on the difference between an adjective and a transitive verb, or an intransitive verb and a transitive one. Regardless of which analysis one prefers, it is the case that the complement functions in the two constructions form a natural class in LMT. OBJθ and OBLΤ appear in the same column of the table, example 63 of chapter 1, that headed by the feature [+r]. Therefore, if this feature is used in the lexical entry of emotion and cognition verbs, we can capture the generalisation that in the two different constructions, these predicates have 195 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 the same semantic argument array. The differences in surface coding are due to other properties of the predicates. I therefore propose the following lexical entries for the two possibilities for the predicate senang 'like': 209. senang, Vtr 'like' < x, y > | [+r] senang, Vitr/A 'like' < x, : y > | [+r] In either case, the first argument will be assigned the feature value [-o] by the default principle. In the first case, the second argument maps to the lowest compatible GF, OBJθ: 210. Ali senang rumah itu. while in the second case it is not a direct argument and therefore can only be realised as an OBLΤ. Following the discussion of section 4.1.2.2, I do not treat the preposition as the predicator in the prepositional phrase, but only as a case marker: 211. Ali senang dengan rumah itu. But this realisation is still compatible with the feature value specified in the lexicon. This generalisation across the two possibilities would not be possible if the feature specified lexically were [+o], which is incompatible with the GF OBLΤ. 4.3.2 P&P analyses - VP shells and internal arguments The previous sections have argued that the bare DP complement of an emotion or cognition verb has syntactic properties which resemble those of a second object, an OBJθ in LFG terms. This grammatical function is accepted as basic in LFG, but other syntactic theories do not accord it the same status. In particular, P&P syntax, since Larson (1988), analyses second objects as the 196 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 result of DP-movement in a particular VP configuration. The aim of this section is to show that such an analysis of the Indonesian data under consideration is implausible. In Larson's analysis of the double object construction (1988, afterwards ODOC), ditransitive verbs project nested VP configurations. The basic structure for a clause containing such a verb is the following: 212. The double object configuration is the result of an operation in the lower VP which ODOC describes as similar to passive in that the specifier or subject position becomes athematic. By Burzio's generalisation (Burzio 1986), the verb therefore is not able to assign case to its complement; that argument moves to the specifier position and the other argument (the theme) is realised as a V' adjunct: 213. In all cases, the verb raises from the lower VP to the higher VP (before moving to a higher functional head if necessary). 197 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 The first problem in applying this type of analysis to the Indonesian data is to decide what the lexical characteristics of the verb should be, that is, what d-structure does it project. If a layered VP is only projected by a verb with two internal arguments, then we must conclude that Indonesian emotion verbs are of this type. Further, we must conclude that in their basic configuration, these verbs project the experiencer argument to a lower structural position than the stimulus. This must be the case as it is the stimulus which will end up as the second object and it must therefore originate as the specifier of the lower VP. Then a passive-like operation must obligatorily apply in the lower VP, and, following this, the experiencer must move to subject position. These instances of DP movement are not the problematic part of the analysis. An unaccusative verb does not assign a thematic role to a subject position, and as noted by Burzio (1986), does not assign case to its object. It is plausible to extend these generalisations to the question of two-place unaccusatives by assuming that such verbs cannot assign case from either of the positions they occupy within VP. Thus, in a structure such as 212, the conditions are met for Larson's PASSIVE operation: no thematic role is assigned to the specifier of the lower VP. This results in a structure like 213, and on the assumption suggested here, the verb in its higher position cannot case-mark the DP in the lower specifier position. Therefore this DP has no option but to move upward again, finally to the case position, specifier of IP. The second object can receive case from the verb if Belletti and Rizzi's (1988) revision of Burzio's generalisation is adopted. Their version of this principle states that a verb which does not assign a thematic role to its subject, cannot assign structural case to an object. Thus, the second object can be assigned inherent case by the verb without affecting the other elements of the analysis28. There is a possible analysis that gives the desired result: that the stimulus argument is a second object derived by movement in the lower VP of a layered structure. However, such an analysis can be ruled out on both empirical and theoretical grounds. Firstly, the discussion of section 4.3.1.1 has shown that either the empirical evidence as to the unaccusative status of these verbs is unreliable, or that there are two classes of verbs. In the first case, as concluded above, there is no reason to accept that any of the Indonesian verbs are two-place unaccusatives. In the second case, the Larsonian analysis would be unable to account for the fact that there are Indonesian emotion and cognition verbs which are not unaccusative on the test proposed (-kan suffixation), but which nevertheless have second arguments which have the same properties as those which are unaccusative on the same test. Theoretically, there is a consensus across all versions of the thematic hierarchy that experiencers are high on the thematic hierarchy, certainly higher than stimuli (or themes)29. Where surface syntactic relations do not reflect this ranking, for example with a verb such as English frighten, various ODOC claims that the lowest V' bar is reanalysed as a V for this, but the details are not relevant here. 28 Dowty's 1991 analysis, mentioned previously, is perhaps the closest to being an exception to this generalisation. 29 198 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 sorts of special pleading have been considered necessary. As previously mentioned, Pesetsky (1995) claims that in such cases the stimulus is not a theme, it does not have the same semantic role as the second argument of fear. He argues for a more fine-grained analysis of semantic roles, specifically that the role causer is important for inverse psych verbs. The question of what is the correct level of detail at which to make generalisations about semantic roles is an important one, but one which is beyond the scope of this study30. Another approach is that of Grimshaw (1990), who accepts the thematic hierarchy, but dissociates thematic and aspectual information. In the case of verbs such as frighten, Grimshaw argues that information from the event structure conflicts with the thematic hierarchy, that is, the stimulus has a causal role and is more prominent in the event that the experiencer, and that this information is treated as more important in argument structure. Note however that Indonesian has only the one type of emotion and cognition verb - there is no syntactic evidence requiring us to propose more fine-grained distinctions here, either in thematic or aspectual structure. The claim that some verbs have two internal arguments and no external argument, that they are two-place unaccusatives, has been made by Belletti and Rizzi (1988) for two of the three classes of psych-verbs in Italian. However, the essential feature of their analysis is that the d-structure for such verbs reflects the thematic prominence of the arguments although the surface relations are reversed (that is, the stimulus is the surface subject)31. For the remaining class of verbs (the temere class), the experiencer is subject and the theme is object. The null hypothesis in this case must be that that if surface thematic relations are as we expect, then so are relations at any other level. Belletti and Rizzi do not question this assumption and argue for an analysis of the temere class which respects it, that is, that the experiencer is an external argument. There is an otherwise plausible analysis in Belletti and Rizzi's scheme which captures the Indonesian facts - that the verbs are two-place unaccusatives which assign case inherently to the theme argument, explaining its second-object properties and also why it is the experiencer which must move - but this is ruled out by the considerations just discussed. 4.3.3 Summary This section has argued that an LFG analysis is possible which captures the fact that the second argument of Indonesian emotion and cognition verbs has properties very similar to those of second objects. It has also been argued that two influential approaches in the P&P framework, those of Larson (1988) and Belletti and Rizzi (1988) are not capable of providing such satisfactory analyses. The following section turns to the question of whether it is a theoretically or empirically adequate conclusion to accept that the arguments Dowty 1991 and Foley and Van Valin 1984 present arguments for only recognising two roles as relevant to syntax. At the other extreme, McRae, Ferretti and Amyote (1997) present psycho-linguistic evidence suggesting that no such generalisations are used in comprehension. 30 Belletti and Rizzi's analysis is thus a P&P reworking of the Relational Grammar inversion analysis of such constructions (Blake 1990: 46-50) 31 199 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 in question are second objects, or whether taking a different view of Indonesian clausal syntax as a whole can offer a better solution. 4.4 Theoretical and empirical problems The previous section was mainly concerned to establish that the technical apparatus of LFG was adequate to present the analysis of Indonesian emotion and cognition verbs which was proposed in section 4.2, and to make the analysis explicit. I now turn to the question of whether this analysis sits comfortably within the broader LFG framework, more specifically, whether the use made of the grammatical function OBJθ is consistent with the theory. I argue in section 4.4.1 that the analysis is consistent with the theory but is also typologically surprising, and that this is contrary to the spirit of LFG. The crucial assumption of the analysis is that the system analysed in chapter 2 is the paradigm for transitive verbs in Indonesian. Section 4.4.2 asks whether emotion and cognition verbs provide empirical evidence for questioning that assumption and provides an initial suggestion of how an alternative analysis of the system might look. 4.4.1 The status of OBJθ in LFG Classic LFG (Kaplan and Bresnan 1982) admitted OBJECT-2 as a primitive grammatical function. The GF was necessary for the analysis of, for example, English double object constructions such as: 214. Mary baked James a cake. But in a theory where the inventory of GFs did not depend on any other theoretical construct, the status of this GF could be marginal; it could be a necessary evil with no theoretical consequences. However, with the introduction of LMT (Bresnan and Kanerva 1989), the GF OBJθ has a rather different theoretical status. The apparatus of LMT predicts that four grammatical functions, and only four, must exist. The four are placed in a markedness hierarchy, but the theoretical apparatus accords them all equal legitimacy. The restricted role played by OBJθ is then somewhat surprising, as is the fact that not all languages have this GF. This point can be made clear by considering the asymmetry between the two markedness relations established by LMT. SUBJ is the unmarked GF with both OBJ and OBLΤ more marked. Typologically, this plausibly translates as the claim that SUBJs appear more freely in a language than the other two GFs, but not as a claim that there are languages which have SUBJs but which lack all other GFs. Any such claim would be incoherent to the extent that SUBJ is a relational notion: it is the grammatical function of the most prominent or most syntactically privileged DP in the clause and this clearly implies that it must be possible for there to be other DPs in the clause with some other grammatical function32. But the relation between OBJ and OBLΤ on the one hand and OBJθ on the other is This argument applies in cases such as Mel'cuk's analysis of Lezgian (1983) as a language without transitive verbs. The non-subject DPs still have to have some grammatical function, in this case oblique. A more complex case is provided by the analysis of Tagalog as having only equational sentences - this is discussed in detail in chapter 5. 32 200 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 different: in this case the increase in markedness can mean that the most marked function is optional. This does not follow from the apparatus of LMT, where the theoretical difference in markedness is the same in the two cases. It might be suggested that the evidence presented here justifies proposing a more fine-grained analysis of non-subject arguments in Indonesian than is allowed by LFG. The syntactic properties of DP stimuli and OBJθ s are not identical, and the considerable overlap might be due only to the fact that both are types of direct argument. There might be more than three types of non-subject argument in the language. However, the LMT offers a strong theory-internal argument against adopting such an analysis. Two binary-valued features allow exactly the four GFs assumed by the theory. and introducing a new GF would disrupt this aspect of the theory. To do so must mean that the two features used thus far do not exhaust the relevant information, and another feature would be required. This in turn would mean that rather than adding one GF, at least two would have to be added, on the assumption that the new feature would be dependent on one of those already assumed. If the new feature was independent of the existing ones, then four new GFs would be added to the system. To my knowledge, there is no empirical support for reducing the constraints on the theory in this way. This consideration suggests that LFG should either seek an explanation of the asymmetry discussed in the previous paragraph, or should seek analyses, such as that proposed here, which accord a greater importance to the OBJθ grammatical function. Against this is the aim of LFG to be what might be called a 'typologically responsible' framework. That is, the framework aims to deal with the full typological range of languages without doing violence to the facts. And the typological facts are that second objects are not used in all languages and that where they do exist their occurrence is limited. There is empirical evidence from a number of languages, some referred to in previous sections, that subject and indirect object, or a nominative and dative case array, is a diathesis that occurs in many languages and that this is associated with predicates of low transitivity (see Blume 1998 for a partial survey). The data under consideration here might be thought to fall under this category, but the match is in fact poor. Indonesian is very similar to English in its treatment of three-place events (modulo the presence of morphological applicative). Recipients and beneficiaries, the prototypical semantic roles which receive dative case in case-marking languages, are coded either as obliques or as direct objects; they are never second objects. Some of the examples of this clause type in other languages are restricted to verbs denoting psychological processes, and these might be considered particularly relevant. But as discussed in previous sections, such verbs have the experiencer in dative case. There is then a double mismatch with the Indonesian data: dative is not the relevant semantic relation and experiencer is not the semantic role which is treated in non-standard fashion. Therefore on typological grounds the proposed analysis of Indonesian emotion and cognition verbs is extremely unusual. The fact that the apparatus of LFG can accommodate the analysis is not a sufficient reason for accepting it without examining alternatives. 201 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 4.4.2 Empirical issues A basic assumption that has guided the discussion in this chapter since section 4.2 is that a two-place verb which does not enter into the system analysed in chapter 2 is not a transitive verb in Indonesian. The analysis of the second argument of emotion and cognition verbs as a second object is a direct result of this assumption. However, in addition to the concerns discussed in the preceding section there are empirical reasons to be uncomfortable with the analysis and to question the assumption. Firstly, the syntactic properties of the arguments in question are not identical to those of second objects. Relative clause constructions with the stimulus argument as head are accepted by all speakers as fully grammatical, as noted by Stevens in 1970 (see example 179 above). On the other hand, relative clauses with second objects as head are impossible for many speakers, are marginal for some speakers, but subject to lexical idiosyncrasy even for them (see example 191). Secondly, the type of clause in which emotion and cognition verbs appear is strikingly similar to the type of clause referred to as a bare verb clause in chapter 2. In this clause type, an unaffixed verb appears and the actor is always subject. In particular, extraction of the second argument is fully grammatical from this type of clause: 215. Inilah buku yang Badu sudah baca this.EMPH book REL Badu PERF read 'This is the book which Badu has read already.' (Voskuil 1996: 189, ex4) On the assumptions adopted to this point, the second argument is an OBJ in one case and an OBJθ in the other case, despite the great similarity in their syntactic properties. I argued in section 2.3.2 that the impossibility of extracting non-subject arguments was characteristic of the type of clausal organization found in Philippine languages, and that some Indonesian clause types showed features of that type of organization. I will now suggest that this line of thinking can be pushed rather further, and that Indonesian should be analysed as having two types of clausal organization available side-by-side. One type has a verb phrase as a major constituent of the clause, whilst the other has a predicate phrase as a major constituent, which may in turn be headed by a verbal root. Instead of taking the ability to appear in the full range of clause types discussed in chapter 2 as criterial for transitive verbs, it is better to take only those clauses with an unprefixed (where prefix includes morphologically attached pronoun) verb and two arguments as transitive verbal constructions. The other clause types in which some verbs appear are then analysed as subject-predicate clauses, in the same way that the clauses with nominal, adjectival and prepositional heads are traditionally analysed. The function of the verbal prefixes can then be analysed as derivational: they allow verbs to head predicate phrases. Emotion and cognition verbs are regular transitive verbs on this account; their oddity is that they do not allow derivation of predicate heads without prior affixation with other morphology. 202 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 4 The division between these two modes of clausal organization correlates with a difference in how arguments are licensed. In addition to changing the status of verb roots, the prefixes meN- and di- must also specify the linking of an argument to the SUBJ function, as argued in chapter 2. This means that LMT does not play a role in the licensing of SUBJs in these clause types. On the other hand, the linking properties of both bare verb clauses and emotion and cognition verb clauses are exactly as would be predicted by LMT: the more thematically prominent argument is linked to the SUBJ function consistently, and, if the new proposal is correct, the non-subject argument is an OBJ as expected. The difference in the properties of non-subject arguments between the two clausal structures, verb phrase and predicate phrase, may also be a result of some difference in how the arguments are licensed. In chapter 5, additional evidence for the dual structure hypothesis is examined, and some problems in stating the hypothesis in the LFG framework are discussed. 203 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 5 Two types of clause structure This chapter examines in more detail the hypothesis introduced at the end of the preceding chapter: that Indonesian has two types of clause structure which exist side by side, but which have different properties. I will take it that the crucial contrasts to be accounted for are the possibilities for extraction of the non-subject argument and the distribution of attached pronouns, both of which depend on the type of verb in the clause. That the two phenomena are closely linked is made clear by the examination of the secondary strategy for relativization in Indonesian, which uses -nya as a resumptive pronoun. This is discussed in section 5.1 and arguments are presented to show that the relativization possibilities of various syntactic positions cannot be derived from the binding theory properties of the pronoun -nya. The distribution of -nya is examined in section 5.2, and I argue that the generalisation which describes that distribution makes crucial reference to the notion of possible predicative elements, with prefixed verbs falling into this class. Section 5.3 presents evidence that this is a plausible approach to the prefixed verbs based on both language-internal evidence and comparative and historical considerations. Section 5.4 argues that the intuitions which are the basis for treating prefixed verbs as different in category to unprefixed verbs cannot be captured by an LFG analysis in any straightforward way, and that the resources of the theory are not adequate to deal with the data examined here. It is not my intention to reformulate the theory, so I offer only very brief suggestions as to how the problem might be overcome. Section 5.5 summarises the conclusions of this chapter, and of the work as a whole. An appendix considers the question of whether Indonesian is a counter-example to the Accessibility Hierarchy of Keenan & Comrie (1977). 5.1 Strategies of relativization In previous chapters, many examples have been given which included relative clauses. All of these have exemplified the major relativization strategy of Indonesian, gapping of the coreferential nominal in the embedded clause: 1. gadisi yang Øi dilihat Umar girl Umar REL US.see 'the girl that was seen by Umar' itu that I have argued that the possibility of a non-subject argument being the gap in such relative clauses is criterial for identifying what I will call the verb phrase type of clause in Indonesian, and this test shows that the bare verb actor subject clause type and the emotion verb clause type with a DP non-subject belong to this type. There is another strategy for forming relative clauses in Indonesian, however, and its distribution reveals additional information about clause structure. 5.1.1 Resumptive pronoun strategy Relative clauses can also be formed with a resumptive pronoun taking the place of the head nominal in the embedded clause. The resumptive 204 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 pronoun is always the attached pronoun -nya, and this strategy can therefore be used for any DP which occupies a place in the relative clause where -nya is possible. This means that possessors, non-subject arguments of meN-prefixed verbs, and the objects of some prepositions can head relative clauses: 2. orangi yang mobilnyai dicuri person car.3 REL US.steal 'the person whose car was stolen' (IRG: 288) 3. [sebuah lagu]i yang barangkali saudara song perhaps brother one.CLASS REL menyukainyai AS.like.APPL.3 'a song which perhaps you will like' (IRG: 289) 4. akan FUT rumahi di yang belakangnyai ada pohon mangga house behind.3 exist tree mango LOC REL 'the house behind which there is a mango tree' (IRG: 290) I have argued in previous chapters that the gapped strategy, although the most common means for forming relative clauses, can only be used in two cases: where the head noun is also subject of the embedded clause, and where the head noun is the object of an unprefixed verb in the embedded clause. These are also positions in which an attached pronoun is impossible. I have demonstrated this for the case of the unprefixed verbs elsewhere and now I will present data showing that subjects are never attached pronouns in Indonesian. 5.1.2 Subject and reduced pronouns There is no typological objection to the possibility that subjects should be reduced pronouns. Many scholars assume that the subject is the most topical entity in a clause (in the sense that it is the typically the most continuous in discourse, not in the sense that it is an example of the LFG discourse function TOPIC) and topical elements can often be expressed in a reduced fashion, with zero anaphora, cliticization or pronominal expression often preferred (Givon 1983). Nor is it true that other languages from the Austronesian family disallow subject reduced pronouns. For example, in Tagalog, pronouns generally cliticize to a position 'after the first daughter of the smallest maximal projection which contains them' (Kroeger 1993: 121), and this is true for both nominative (subject) clitics and others1: 5. Dito siya magtatayo ng=bahay here 3SG.NOM FUT.AV.build GEN=house 'Here he will build a house.' (Kroeger 1993: 122) Glossing as in the source: AV - active voice, OV - objective voice. The = sign indicates that the case particles form a phonological unit with the following word, the form kay in example 6 is a suppletive form used with proper names. 1 205 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 6. [Para kay=Pedro] ko binili ang=lauran for 1SG.GEN PERF.buy.OV NOM=toy DAT=Pedro 'For Pedro I bought the toy.' (Kroeger 1993: 122) The correct analysis of the nominative argument in Tagalog is the subject of considerable debate (Cena 1995, Kroeger 1993, Schachter 1976, 1995, Sells 1998, among others). There is a consensus however that the argument glossed here as NOM(inative) is the most syntactically privileged, and it is clear that with respect to cliticization, it is treated in the same fashion as the nonsubject argument of a transitive clause. Other Austronesian languages for which the grammatical function subject may not be applicable still show this sort of symmetry with respect to the two arguments of transitive verbs. Acehnese (Northern Sumatra), which Durie (1985) argues has no subject function at all, uses clitics to code both actor and undergoer of a transitive clause. The actor clitic is obligatory and the undergoer clitic is optional2, in each case the clitic is capable of being the sole exponent of the argument in the clause: 7. ji= kap =keuh 3 bite 2 'It'll bite you!' (Durie 1985: 47, ex4-3) Similar facts can be observed in the Menó-Mené variety of Sasak (SouthCentral Lombok). It is not clear whether this variety can usefully be analysed in terms of subject and object functions (Austin 1997b, Kroon 1998), but both arguments of a transitive verb can be represented by clitic pronouns. As in Acehnese, positional coding distinguishes the two possibilities: actor clitics attach to auxiliaries and undergoer clitics attach to verbs: 8. Yaq-n bèng-k hadiah-nò give-1SG present-that FUT-3 'He will give me that present.' Again, a clitic pronoun can be the only exponent of the semantic role in the clause. Indonesian completely lacks the option of having the subject coded as a reduced pronoun. The reduced pronouns seen in some Pro-V clauses have already been treated in detail in chapter 2, and they are not subjects (see also section 1.2.2). The other set of reduced pronouns which exist are -ku, -mu and -nya. These attach to the right edge of a preceding word. As subjects are very often the first element in the clause, linear order might be assumed to prevent this type of expression for subjects. But even when some element precedes the subject, it is not possible for that element to host an enclitic3: This is true of the dialect from North Aceh which Durie describes. Southern dialects omit the undergoer clitics altogether (Bukhari Daud, p.c.) 2 3 This is possible in Sasak, for example. 206 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 9. *Ali memikirkan bahwanya Ali that.3 AS.think.APPL (FOR: 'Ali thinks that he is sick.') sakit sick The possibility of a reduced pronoun attaching to the right of an auxiliary also does not exist in Indonesian: 10. *Akannya pulang go.home FUT.3 (FOR: 'He will go home.') The only type of structure which seems to contradict the claim made here is the 'raising-to-object' type: 11. Siti menganggap dia 3SG Siti AS.consider 'Siti considers him foolish.' bodoh foolish In this type of structure, the subject of the embedded clause can be encliticized to the matrix verb: 12. Siti menganggapnya Siti AS.consider.3 'Siti considers him foolish.' bodoh foolish But there is good evidence in such cases that the subject of the embedded clause is the non-subject argument of the matrix verb. It can be made subject in a passive di-V clause: 13. Dia dianggap bodoh oleh 3SG US.consider foolish by 'He is considered foolish by Siti.' Siti Siti The generalisation can therefore be maintained: subjects in Indonesian are never reduced pronouns. The positions which can be gapped in relative clause formation are in complementary distribution with the positions where a resumptive pronoun can occur (there are also positions where neither is possible). 5.1.3 Complementary distribution of relativization strategies The data presented above show that there is a complementary distribution between the positions in which reduced pronouns can occur and the positions from which gapped relativization is possible. This fact might lead one to search for an explanation for the restrictions on gapped relativization in terms of the binding properties of either the resumptive pronoun or of the crucial positions. This section will present evidence that no such explanation is possible, as well as briefly discussing one analysis along those lines. 207 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 5.1.3.1 TOPIC-COMMENT STRUCTURES The resumptive pronoun strategy for relativization is generally agreed to be related to the possibility of Topic-Comment structures of the following type (Alsagoff 1992: chapter 7, IRG: 288-291): 14. [Orang itu]i, saya melihat person that 1SG AS.see 'That person, I saw them.' -nyai/diai 3SG Fundamental assumptions of all versions of binding theory tell us that the topicalized DP in such structures must be outside the binding domain that contains the resumptive pronoun. In terms of Chomsky's formulation (1981), this is a Principle B effect: a pronoun cannot be coreferential with a ccommanding nominal in the relevant domain. The topicalized element and the pronoun co-refer and, equating leftward position with height in the tree, the pronoun is c-commanded by the topicalized element. Therefore, they cannot be in the same binding domain. The same argument can be made in the LFG version of binding theory, where the functional nucleus is taken as the relevant domain (Bresnan 2001a: chapter 10). Where a possible antecedent of either pronoun occurs within the relevant domain, coreference with the pronoun is not possible: 15. melihat [Orang itu]i, Alij person that 1SG AS.see 'That person, Ali saw them.' -nyai/*j/diai/*j 3SG When the free pronouns in examples 14 and 15 are replaced by -nya, precisely similar arguments apply in that case also. But in topic-comment structures, the distribution of dia is wider than that of -nya. The free pronoun can appear in subject position which is impossible for -nya: 16. 16' [Anak itu]i, diai mencubit doktor *Anak itu, nya mencubit doktor child that 3SG AS.pinch doctor 'The child, she pinched the doctor.' (Adapted from Alsagoff 1992: 228, ex22a) itu itu that According to Alsagoff (1992: 228), the resumptive pronoun can also appear in second object position. Such examples are considered ungrammatical by my consultants; whether this difference is due to variation between Indonesian and Malaysian, or whether it should be attributed to some other reason, is not clear. But at least some speakers of Malaysian allow examples such as the following, and again dia has a wider distribution than -nya: 17. 17' Mariami, Ali menawarkan doktor itu diai *Mariam Ali menawarkan doktor itu nya doctor that 3SG Mariam Ali AS.offer 'Mariam, Ali offered her to the doctor.' (Alsagoff 1992: 229, ex25a) 208 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 The binding properties of dia have not changed between these various examples: coreference with the topic is perfectly acceptable in all of these cases. But examples 14 and 15 showed that the binding conditions on dia and -nya must be the same. Therefore, the impossibility of -nya in 16' and 17' cannot be due to any binding theory violation, and must be due to some restriction on possible hosts. This might be phonological, or syntactic. Additional evidence comes from the fact that only some prepositions allow an enclitic pronoun as their object. There is no obvious semantic generalisation about which prepositions have this property and which do not, nor is there an obvious phonological generalisation. The two monosyllabic prepositions di and ke do not allow -nya as their object, but monosyllabic nouns are fully grammatical with -nya as possessor, therefore it seems unlikely that this could be a relevant factor. I will therefore treat the property as a lexical idiosyncrasy4. Regardless of whether the preposition can host a clitic, dia can always follow it as the resumptive pronoun in a topic-comment structure: 18. Lisai, semua orang ingin bertemu Lisa all person wish meet 'Lisa, everyone wants to meet her.' 19. Lisai, semua orang ingin berbicara Lisa all person wish talk 'Lisa, everyone wants to talk about her.' dengan with diai 3SG tentang about diai 3SG The enclitic -nya is only possible when the preposition has the lexical property of allowing it: 18' Lisai, semua orang ingin bertemu Lisa all person wish meet 'Lisa, everyone wants to meet her.' 19' Lisai, Lisa semua all orang person ingin wish berbicara talk dengannyai with tentangnyai about Again there is a restriction on possible hosts for -nya, but in this case it is due to some lexical property of prepositions. The binding properties of the two forms of the pronoun are identical, therefore they cannot be the explanation for the impossibility of -nya in one clause but not in the other. In relative clauses, the resumptive pronoun can only be -nya. But the properties of such structures as far as binding theory is concerned are effectively identical to the topic-comment structure. The arguments given in the preceding paragraph apply in exactly the same fashion: only -nya can appear in relative clauses, but this cannot be due to the binding properties of such structures. I have not investigated the question in detail, and I certainly do not rule out the possibility that further research might reveal some interesting pattern. This issue is discussed further in section 5.2. 4 209 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 5.1.3.2 VOSKUIL 1996 Voskuil (1996: chapter 8, afterwards V) offers an account of the impossibility of A-bar movement from the position of the non-subject argument of prefixed verbs in Indonesian in the P&P framework. The first crucial assumption that V makes is that verb prefixes license pro, a zero pronoun, as the object of the verb they are attached to. Both prefixed verbs and bare verbs allow zero anaphora of their non-subject arguments, and V claims that a different empty category is involved in the two cases. Following a prefixed verb, pro will be left, but this is not possible for a bare verb. In that case, V assumes that a zero operator is moved to a topic position and that this binds a trace in the position of the non-subject argument: 20. Lupus langsung menghampiri pro Lupus immediately AS.approach 'Lupus immediately approached (him).' (V: 193, ex29) 21. Opi mak sudah beli ti 1SG buy PERF 'I already bought (some).' (V: 193, ex30) Some of the evidence supporting this assumption has been discussed in section 2.2.2.2 and I will not repeat it here. V's second crucial assumption is the following principle: 22. An empty category is interpreted as pro if morphological conditions for pro are met, irrespective of derivational history. This principle is motivated with arguments from Italian and Irish. Irish has anti-agreement effects: an overt subject and agreement on the verb cannot cooccur. When a subject is extracted, agreement cannot occur on the verb of the clause from which the extraction has occurred5. V reasons that in such cases, the presence of agreement would license pro and therefore force interpretation of the empty category as pro, which cannot be the foot of a chain. The principle 22 therefore predicts that a gap in the position of the non-subject argument of a meN-prefixed verb cannot be interpreted as a trace, it cannot be the result of movement. This leaves open the possibility that it can be interpreted as pro, but coreferential with a DP outside the binding domain of pro. Nothing in V's account, or in the general principles which he assumes, rules out the possibility of pro being a null resumptive pronoun in a structure such as that seen in example 3 (repeated here): 3. 5 [sebuah lagu]i yang barangkali saudara one.CLASS song perhaps brother REL menyukainyai AS.like.APPL.3 'a song which perhaps you will like' (IRG: 289) akan FUT The Irish data are discussed in detail and given an LFG analysis in Andrews (1990). 210 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 3'. *[sebuah lagu]i yang barangkali saudara one.CLASS song perhaps brother REL menyukai proi AS.like.APPL.3 'a song which perhaps you will like' (IRG: 289) akan FUT pro has the feature matrix [-anaphor, +pronominal], that is, it is a pronoun and is predicted to have the same distribution as an overt pronoun (Chomsky & Lasnik 1995: 36). The structure shown in 3' should therefore be possible, but it is not. I will give only a brief sketch of V's analysis of this problem, because the analysis is based on a significant empirical error. V claims that examples like 3 are ungrammatical, contradicting the sources which I have consulted (Alsagoff 1992: chapter 7, IRG: 281) as well as native speaker judgments. The question V attempts to answer is why 3 is not possible, the answer to this will rule out 3' and complete the account of the ungrammaticality of the extraction of non-subject arguments. V's account treats resumptive pronouns as variables and claims that the binding relation between a variable and some other element must be due to movement, except when other principles prohibit movement from the position of the variable. In the case of structures like 3, no principle bars movement from the non-subject position, the equivalent in English is well-formed. Therefore, the variable must be derived from movement, and 3 is not well-formed6. The empirical error in V's account unfortunately distracts his attention from the real question: why 3 is grammatical, but any similar structure without the resumptive pronoun is not. 5.1.3.3 SUMMARY The preceding sections have shown that the distribution of the attached pronoun -nya and the distribution of the two strategies for forming relative clauses in Indonesian are intimately linked. I have also argued that no explanation for this distribution can be found in the binding theoretic properties of the pronouns involved. The question that therefore needs to be answered is: what factors do constrain the distribution of -nya, and hence of relativization strategies? I take up this question in the following section. The attentive reader may have noted in the discussion of the secondary relativization strategy that the position of a second object is not a position in which -nya can occur. It is also not a position from which gapped extraction is possible, with the exception noted in section 4.2.3 (example 191). But possessors and objects of (some) prepositions can be relativized with a resumptive pronoun. These facts suggest that Indonesian may provide a counter-example to one of the universals proposed by Keenan & Comrie (1977), that a relativization strategy should apply to a continuous segment of the hierarchy of syntactic functions. This issue is taken up in the Appendix to this chapter. V's account is less stipulative than might appear from this summary. But in view of the empirical problem, giving the details seems pointless. 6 211 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 5.2 The distribution of -nya The distribution of -nya (and the other two enclitic pronouns) has been discussed briefly in section 5.1.3.1. Here I analyse this question more carefully, and establish a generalisation as to the positions in which enclitic pronouns can occur. Three positions are possible: after prefixed verbs, after some prepositions, and as possessor in a DP. I will discuss them in this order, which reflects the increasing complexity of the issues involved. The case of prefixed verbs has been discussed in detail in chapter 2. All that is necessary to the present discussion is to recall that an enclitic can encode the nonsubject argument of a prefixed verb, and that its syntactic status is different in the case of the two prefixes. The pronoun is a true clitic in the case of meNprefixed verbs, and I assume that it is generated as sister of the head in cstructure. With a di-prefixed verb, the pronoun is an affix attached in the lexicon. This fact already suggests that a generalisation which covers all cases will have to be stated in terms of f-structures. Only a subset of prepositions can have enclitics as their complements. This group includes the following prepositions: 23. pada kepada7 daripada bagi buat untuk bersama dengan terhadap oleh dari at, to to than, from for for for together with with towards by than, from A text example of a preposition hosting an enclitic is the following: 24. Akan kubuatkan susu untukmu 1SG.make milk for.2 FUT 'I will prepare some milk for you.' (SDM: 152) The group of prepositions listed in 23 does not include the prototypical locative preposition, di8. However, enclitics do occur in some locative expressions, for example: 25. di atasnya top.3 LOC 'on top of him/her/it/them' Expressions of this type are best analysed as preposition plus locative noun sequences, rather than as compound prepositions; bracketed as in 26 not 27: 7 kepada is typically used when the complement is a person. 8 This preposition also does not occur with emotion and cognition verbs (see section 4.1.2.2). 212 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 26. 27. [PP [P di ] [DP atasnya]] [PP [P di atas] [DP -nya]] The enclitic in such cases is then an example of the possessive type, and this analysis can be maintained even when di is omitted in informal Indonesian, e.g. atasnya 'on top of it'. It is not the case, however, that there is a neat divide between static locative prepositions and all others with respect to enclitics. For example, tentang 'about, concerning' does not host clitics. The prepositions used to code recipients and beneficiaries are all present in 27. As discussed in section 2.2.1.3, Indonesian pronouns have only human reference for conservative speakers, and historically, this generalisation has covered -nya also. It is possible, then, that the prepositions which allow enclitics are those which are more likely to appear with human complements, and those which do not allow enclitics are those which have prototypically non-human complements, such as di. Aristar (1997) argues analogously that, in casemarking languages, the non-grammatical cases include in their definition some specification of their typical argument. I will assume without argument that prepositions in Indonesian are lexically specified to take a DP complement, and that this is the sister of the P head in c-structure. Enclitics are hosted very commonly by nouns, in which case they code possession of the head noun: 28. Dan kamu akan menyesalinya seumur se.age and 2SG FUT AS.regret.3 'And you will regret it all your life!' (SDM: 125) hidupmu! life.2 The structure of DP in Indonesian is significantly different from that in English as far as possessives are concerned. Common analyses for English have the possessive as the head of D (see e.g. Abney 1987), or more conservatively, as the specifier of DP. The possessive in Indonesian is not a specifier or the head of D. Unlike English, but like for example Italian, possessives and determiners can co-occur, assuming the argument given in section 1.2.3 that quantifiers are exponents of D in Indonesian: 29. Kelima anaknya pandai child.3 clever ORD.five 'All five of her children are clever.' (IRG: 133) 30. Dua pertiga orang Indonesia tinggal two person Indonesia live FRAC.three Jawa Java 'Two thirds of Indonesians live in Java' (IRG: 133) di LOC pulau island Possessives are not specifiers in NP either; they follow the head but are not always final in the constituent. They can be followed by demonstratives: 31. satelit kita itu satellite 1PL.INCL that 'that satellite of ours' (McKay) 213 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 32. negeri saya ini country 1SG this 'this country of mine' (McKay) by relative clauses: 33. kepala Ardi yang tadi ditahan oleh sandaran before di.support by support head Ardi REL kepala kursi head chair 'Ardi's head that was previously supported by the head rest of the chair' (SDM: 202) and by some other adjuncts. Possessors are not necessarily the modifier closest to the head noun. Adjectives and modifying nouns can intervene: 34. guru bahasa saya teacher language 1SG 'my language teacher' 35. sikap tegas bapak attitude steady father 'father's firm attitude' and the possessor can be realised as a bound pronoun in this position: 36. guru bahasaku teacher language.1SG 'my language teacher' When a modifying noun intervenes between the head noun and the possessor, there is a potential ambiguity and this is not dependent on whether or not a bound pronoun is used. Thus both example 34 and example 36 have a (marginal) reading 'teacher of my language'. The facts described above make it clear that in Indonesian DPs, possessors are part of the NP complement of the functional head. The fact that constituents such as relative clauses, unambiguously part of the projection of N, also follow possessors shows that possessors are not specifiers within NP, they must be either complements or adjuncts under an N' projection. Given that they are not obligatorily closest to the head noun, and can be preceded by constituents which are certainly not complements (adjectives), two structures seem possible: either modifying nouns and adjectives always form compound nouns in Indonesian, or N' is a freely recursive category in Indonesian. These two structures are illustrated in the following trees9: I assume that demonstratives are specifiers of NP, but not determiners. Nothing depends on this assumption. 9 214 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 37.a 38. b buku baru saya itu book new 1SG that 'that new book of mine' The first alternative is an extremely stipulative solution; allowing it would considerably weaken X-bar theory as applied to Indonesian. Therefore, the second structure is preferable. This result means that the three cases in which enclitics are allowed cannot be brought under a single generalisation at c-structure, even if we ignore di-prefixed verbs. For meN-prefixed verbs and for prepositions, the clitic is the immediate complement of the head, but this is not the case for nouns. The structure argued for above places possessors in a position typical of adjuncts, under an X' but not under the lowest X'. If a unifying generalisation exists, it must therefore be sought in f-structure. At first sight, this approach does not seem much more promising. For prefixed verbs (both types now), the attached pronoun encodes the nonsubject argument, the second argument. For prepositions, the enclitic encodes the complement, possibly the only lexically specified argument. And for nouns, the enclitic encodes an optional argument. However, this description is overly reliant on familiar concepts from English grammar. Taking closer account of the facts of Indonesian grammar gives a more revealing picture. Section 4.1.3 presented data showing that all major categories in Indonesian can be the main predicate of a clause, and this fact is the clue to the generalisation which describes the distribution of enclitics: an enclitic can only be the non-subject argument of a (potential) predicate. This statement covers the case of di-prefixed verbs also, as they have the actor as non-subject argument in f-structure. The line of argument being developed here requires a corollary of this generalisation: unprefixed verbs cannot be predicates in the same sense, as they have non-subject arguments which can never be enclitics. I now consider the question of how this contrast might be understood as the outcome of other factors in the grammar of Indonesian. 5.3 Potentially predicative elements and arguments The previous section has argued that the correct characterisation of the positioning of attached pronouns in Indonesian makes reference to the non- 215 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 subject argument of potential predicates. I will refer to such elements as being potentially predicative. If these potential predicates have a non-subject argument, the implication is that they also have a subject argument. But it is rather implausible that, for example, all Indonesian nouns have a subject argument as a lexical property. This section will argue that this objection can be met when we consider more closely the properties of potentially predicative elements, and that there is a significant similarity between prefixed verbs and other potential predicates in the relation between predicate and subject. 5.3.1 Semantic roles Verbs assign idiosyncratic semantic roles to their arguments. Thus the subject of the English verb fall is understood to move (involuntarily) from a higher point to a lower point during the falling event, while the subject of rise moves in the opposite direction and may do so voluntarily. Similarly the nonsubject arguments of two verbs such as cut and weigh are understood very differently. Linguists use thematic roles as convenient generalisations over these individual roles, but the semantics that verbs assign to arguments are more complex. This is not true in the same fashion for non-verbal predicates. Such predicates may assign a variety of roles to an argument internal to the predicate, sometimes but not always indicated by preposition choice in English: consider the difference between a student of linguistics and an opponent of linguistics. But the roles assigned to the subject of predication are much more limited. Lyons (1977: 469) lists four non-verbal sentence schemata covering equative, ascriptive, locative and possessive predications, and other classifications in the literature are similar (e.g. Hengeveld 1992). This classification, or a very similar one, can be justified on syntactic grounds in various languages and the semantic distinctions are intuitively clear. However, it is possible to make a semantic generalisation over these four types: all of them can be seen as ascribing a property to the subject of predication. Besides simple properties, there is the property of being identical to something else, the property of being in a certain location and the property of being possessed by someone. In this sense, the subject of all these predication types has only a single semantic role, that of bearer of a property, and the predicate itself has only one semantic function, to specify the property10. Notice also that such structures do not allow the possibility of switching the direction of predication, except in the special case of identity. While it is common for languages to have syntactic and even occasionally lexical resources which allow an event to be represented from different viewpoints, this is not generally possible for non-verbal predications except by some type of paraphrase. This is, of course, linked to the fact that languages do not have extensive resources for this type of predication. Either, as in Indonesian, there is no copula in most such predications, or there are limited I am aware that I am sidestepping a number of ancient and contentious philosophical debates at this point. The reader is referred to Seuren 1998: 121-133 and Moro 1997: 248261 for some discussion. 10 216 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 lexical resources. Hengeveld (1992) reports languages with several copulas and what he terms semi-copulas, but these are used for different types of predication, not for altering the directionality of the relationship. Thus, I gave Helen some money, and Helen received some money from me are representations of the same event using different lexical predicates, but to change the viewpoint with a predication such as I was in London, the only option is rather stilted paraphrase London was my location. The essential point here is that the semantic role of the subject of such predications is fixed, or at least limited to a very small set of possibilities. 5.3.2 Grammatical functions The fact that all major categories in Indonesian can head a predication raises questions about the nature of the grammatical function SUBJ in the language. Do all categories include in their lexical entries the information that SUBJ is a GF that they can assign? There is some relevant empirical evidence, but it is not conclusive. There is a strong tendency for adjectives which modify nouns to be introduced by yang. That is, many adjectival modifiers are formally relative clauses, with the modified noun as the gapped SUBJ: 39. Kemudian dipukulnya bola itu sekuat-kuatnya. after ball that se. force.DUP.3 US.hit.3 Servis yang sempurna service REL perfect 'Then she struck the ball forcefully. A perfect service.' (SDM: 5) In some cases, for example when more than one adjective is used, the relative clause structure is obligatory: 40. Dari Amsterdam Ardi membawa Marisa ke Paris. Marisa to Paris from Amsterdam Ardi AS.take Kota cantik yang hangat (...) yang ramah dan warm friendly and city beautiful REL REL eksotis exotic 'From Amsterdam, Ardi took Marisa to Paris. The beautiful, warm city ... friendly and exotic.' (SDM: 50) Thus adjectives have at least a tendency to appear in predicative structures. This is certainly related to the blurring of the distinction between adjectives and intransitive verbs in Indonesian, discussed in section 4.1.3.2. Prepositional phrases often modify nouns also, and relative clause structures like the adjectival ones are possible: 41. anak yang seperti Tini child like Tini REL 'children who are like Tini' (IRG: 287) but this is less common than with adjectives. Most such modifiers of nouns are simple PPs: 217 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 42. masih adakah tempat untuk dendam di hatimu? still exist.EMPH place for revenge liver11.2 LOC 'Is there still a place for revenge in your heart?' (SDM: 55) 43. Marisa menatap hasil pemeriksaan laboratorium Marisa AS.observe result examination laboratory untuk urine-nya itu dengan kesal for urine.3 that with cross 'Marisa stared at the result of her urine test crossly.' (SDM: 128) Treating this type of structure as syntactically consisting of a subject and a predicate would be very problematic. There are clear reasons against such an analysis both from the point of view of constituent structure (the PP is a modifier, not in a specifier-head relation with the noun it modifies) and from the point of view of functional structure (the modified noun would have either have two functions assigned it, or a controlled position would have to be postulated). The last example, example 43, also raises another problem in the analysis of PPs as having subjects: they can be modifiers of events as well as of entities as the PP dengan kesal is in this example. Nouns only have a subject in predicative structures. The kind of arguments that have been made in respect of English to treat the possessor position as a subject position (M.Anderson 1983, Grimshaw 1990, Stowell 1983 among others) are not relevant to Indonesian. As discussed in chapter 1 and in section 5.2 above, possessors in Indonesian are arguments internal to the projection of the head noun, and not specifiers of any projection. There is no reason to assume that nouns have SUBJ as an assignable function in their lexical entries. On the basis of the evidence discussed above, there is also little reason to assume that prepositions have subjects either, and consistency would then suggest that adjectives ought to be treated in the same way. I take it that this conclusion is related to the position argued for in the previous section: the lack of a SUBJ function is related to the lack of a contentful semantic role for that SUBJ to be linked to. I argued in section 5.2 above that the possible positions of attached pronouns should be understood in terms of elements that are potentially predicative, and that this should include prefixed verbs. This would imply that clauses containing prefixed verbs also have a subject-predicate structure and that such verbs do not have a SUBJ in their lexical forms. This is an intuitively implausible idea: how can it be that a verb does not always have a SUBJ in its argument list? Nevertheless, I will suggest that there are reasons for adopting such an analysis. The most important reason for adopting the suggestion is the very striking parallel between prefixed verbs and the other predicative categories with respect to the assignment of a semantic role to the subject of the predication. I argued above that the semantic role assigned to the subject of non-verbal predications is predictable on general grounds, and there is 11 Indonesian treats the liver as the seat of the emotions. 218 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 therefore no work to be done by any sort of mapping theory. The assignment of a semantic role to the subject of prefixed verbs is accomplished in a similar fashion if the account of the verb prefixes presented in chapter 2 is accepted. That proposal has the consequence that mapping theory is irrelevant for such verbs: the mapping to subject is directly specified by the prefix. The account of chapter 2 will need some revision in light of the proposal presented here, but this crucial insight will be maintained. Under the current proposal, the prefixes meN- and di- are derivational. Their effect is to alter the attributes of the verb root so that its maximal projection will be a predicate rather than a verb phrase. In chapter 1, I claimed that all morphological processes in Indonesian were derivational, therefore this analysis has the advantage of including the verb prefixes in this generalisation. The insight that the SUBJ function of prefixed verbs is not assigned by the operation of LMT can be incorporated in two ways. One possibility is that the verb prefixes have the effect of allowing the maximal projection of the verb to be interpreted as an attribute of the SUBJ. The predicate would have a semantic role Attributee to be associated with the SUBJ as suggested for nonverbal predicates previously. This would mean that the literal reading of a clause such as Saya melihat gadis itu would be along the lines of 'I have the attribute of being the seer of the girl'. This possibility would give a consistent account of all the predicate types, and it also fits into a comparative and historical perspective on Austronesian syntax which will be discussed in section 5.4.3. But this analysis is very close to the nominalisation analysis discussed in section 2.3.3.1, and there are good reasons to reject that analysis for Indonesian. There are actor and undergoer nominalizations available, distinct from the prefixed verb forms (although the morphology is related in the case of the actor forms): 44. Root tatar suruh Gloss upgrade order A-nominal penatar penyuruh Gloss trainer one who gives orders U-nominal petatar pesuruh Gloss trainee messenger , one who takes orders And although the non-subject argument of a prefixed verb has the same coding properties as the possessor of a noun (i.e. enclitics are possible), their behavioural properties are different. The preferable alternative is that the verb prefixes remove an argument from a-structure and thus from the operations of LMT. The semantic role associated with the argument is still available in the lexical entry of the verb and can be semantically identified with the SUBJ in a subject-predicate structure. The lexical entry of the prefixes meN- and diwould be the following: 45. 46. meN-, prefix, __V < x, y ...>, [SPEC ARG di-, prefix, __V < x, y...> [SPEC ARG y] x] and the morphological rule which creates prefixed verbs might be: 219 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 47. Verb prefixing Prefix + V < (a), [SPEC ARG], (b) > ⇒ V [+pred] < (a),(b) >, θj | θj For each prefix, an argument is specified as the argument targeted by the operation of the morphological rule. The rule removes that argument from the a-structure but leaves the semantic role associated with it available for semantic interpretation. Similar changes would be required for the morphological rule for Pro-V clauses (chapter 2, example 152); given the speculative nature of the current discussion, I do not specify this change here. Aside from the issue of semantic roles, the analysis of prefixed verbs as potentially predicative elements makes a prediction about the syntactic properties of such words which is supported by evidence. The prediction is that such words will be able to head nonpredicative constituents without a SUBJ. The data which supports this prediction is discussed in the following sections. 5.3.3 Nonpredicative uses of prefixed verbs The three categories which can head non-verbal predicates, nouns (projecting to DP), adjectives and prepositions, also can all head nonpredicative constituents. The fact that these usages are common, even basic, and do not require the assignment of a SUBJ function (except possibly in the case of adjectives) was taken previously as an argument in favour of the idea that potentially predicative elements need not have a SUBJ in the argument list of their lexical entries. If prefixed verbs are potentially predicative elements also and need not have a SUBJ as a lexical property, then it can be predicted that they should also be able to appear in positions other than that of predicate. This is indeed the case, and in the contexts which might be analysed in this way, there is a contrast in acceptability between prefixed verbs and unprefixed verbs which is in line with the account offered here. Voskuil (1996: 206-208) identifies constructions involving what he calls three different verb modes: infinitive, present participle and gerund. This classification, on the basis of the data Voskuil presents, is due to the nature of the translations of the different possibilities rather than to any clear differences in the Indonesian12. IRG (302-308) treats the topic under two heads: predicate nominalisation and simple nominal clauses, which seem to correspond respectively to a subclass of Voskuil's gerunds, and everything else. Sneddon's predicate nominalizations have different characteristics from the other types, and I discuss them last. Simple nominal clauses consist of the maximal projection of a prefixed verb or an intransitive verb: Voskuil also cites another work (Lappia, M. & J.E.Voskuil (1992) Nominal infinitivals. In S.Barbiers & C. Levelt (ed) Proceedings of the Third Leiden Conference of Junior Linguists) as giving a more detailed treatment. This work was not available to me. 12 220 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 48. Mencari pekerjaan di kota tidak begitu work city like.that AS.find LOC NEG 'Finding work in the city is not very easy.' (IRG: 307) mudah easy 49. Dipukul Yanto sungguh tidak enak! Yanto surely nice US.hit NEG 'Being beaten by Yanto is surely not nice.' (Voskuil 1996: 207, ex17) Such nominalizations do not have the same distribution as DPs, although the overlap is considerable. In the above examples, the nominalised clause is the subject of an adjectival predicate. They can also be the subject of a verb: 50. Membaca buku menambahkan pengetahuan book AS.increase knowledge AS.read 'Reading books increases one's general knowledge.' (Voskuil 1996: 207, ex16) umum common or the complement of a preposition: 51. Dengan mengucapkan terima-kasih, dia meninggalkan with thank.you 3SG AS.leave AS.say ruang itu room that 'Saying thank you, he left the room.' (Voskuil 1996: 206, ex7) Such clauses do not seem to be possible as non-subject arguments of a verb. Generally, substituting a bare verb for the prefixed form results in unacceptability: 52. *Cari pekerjaan 53. *Baca 54. *Dengan ucapkan ruang itu buku di kota tidak menambahkan begitu pengetahuan terima-kasih, dia mudah umum meninggalkan Voskuil (1996: 206) claims that, in some contexts and for some verbs, a bare verb form is marginally acceptable, and my consultants allow the possibility of such structures in very informal speech13. Overall, bare verb forms are marginal in these structures, and the difference in grammaticality is definitely in the predicted direction. Familiar analyses of such structures in English appeal to the existence of a null pronoun in subject position (P&P) or to anaphoric control of the SUBJ function (LFG). Such accounts depend on the fact that the verb forms used have distinct properties, traditionally summarised by the term non-finite. But there is no distinction in verb form which can be appealed to in Indonesian in the same way: exactly the same prefixed verb forms appear in the 'non-finite' contexts as appear as the predicate of free clauses. Any analysis which Example 54 is judged unacceptable under any circumstances, for reasons to do with register. 13 221 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 postulates a finite/non-finite distinction in Indonesian introduces a theoretical assumption not based on the facts of the language. And such an analysis still has to explain the ungrammaticality of bare verb forms in these contexts: if a subject is projected either in syntax (null pronoun) or in f-structure (anaphoric control), why can it satisfy the properties of one verb form but not the other? The other contrasts between the properties of prefixed and unprefixed verbs which have been discussed to this point referred to properties of non-subject arguments. No data examined previously has suggested that the subjects of the two types of verb differ in any way, therefore this contrast is surprising. Additionally, one of the facts about English that has to be accounted for is that in some non-finite contexts (before -ing verbs) a subject is possible. However, in Indonesian, a subject is never possible in these constructions. These facts support the view that the grammatical function SUBJ is not present at all in these cases. The analysis of prefixed verbs developed here can provide a straightforward analysis of these constructions. A prefixed verb is a potentially predicative category, and it need not have a SUBJ in its lexical entry, it can therefore appear without a SUBJ. Such verbs nevertheless carry semantic information about a potential subject, and that information is used in interpreting these constructions, even though it is not represented syntactically. Note that this account implies that verb roots have a logical subject argument specified in such a way as to ensure that it is mapped to SUBJ, but that prefixation removes this and leaves the possibility (actually high probability) of a SUBJ being added back to the argument list by lexical redundancy rule. This is admittedly messy. An alternative analysis would use a category-changing rule to make a prefixed VP into a DP, with an additional rule required to state that the SUBJ of the verb need not (cannot) appear in such a case. This solution is also messy, and the first alternative at least has the advantage that all the processes assumed to affect potentially predicative words are confined to the lexicon. The difference between the structures just discussed and what Sneddon calls predicate nominalizations is that the subject is expressed in this latter type. Only adjectival clauses, clauses with intransitive verbs, and clauses with di-prefixed verbs can be nominalised in this way, in which -nya is added to the predicate head to make it a noun and the subject follows it as a possessor: 55. Lalu lintas menjadi macet karena ditutupnya stuck because us.close.3 traffic AS.become beberapa jalan several road 'Traffic became jammed because of the closure of several roads.' (IRG: 302) This type of nominalisation raises interesting questions which are not relevant to the present discussion, but they will be discussed briefly in section 5.4.3. 5.3.4 Overt copulas Indonesian has two elements which can be analysed as copulas: adalah and ialah (IRG: 237-238). Both words include the foregrounding particle (IRG: 222 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 261-263) or focus marker (Alsagoff 1992: 114-117) -lah, the other morpheme is the existential verb ada in one case and a third person pronoun in the other. These words can be optionally inserted between the subject and the predicate in a non-verbal predication (ialah can only follow a third person subject), and therefore the account developed here would predict that prefixed verbs could follow these words also. This is possible, and therefore the data give some support to my hypothesis. However, in the case of nominal, adjectival and prepositional predicates, the copulas do not alter the meaning: 56. Soal ini (adalah) penting problem this COP important 'This problem is very important.' sekali very The appearance of the copula in such clauses is generally to make the structure unambiguous, particularly when a nominal clause has a subject with dependent nouns. The copula then clarifies the break between the two constituents. A copula is only possible before a prefixed verb where that verb is to be understood as a nominalisation: 57. Tugas saya adalah menjual koran task 1SG newspaper COP AS.sell 'My job is selling newspapers.' (IRG: 308) The nature of such structures makes it almost impossible to have a minimal pair of examples. However, comparison of example 57 and the following gives some idea of the meaning difference which comes with the copula: 58. Saya menjual koran 1SG newspapers AS.sell 'I sell newspapers.' If prefixed verbs and the other possible predicative elements are as similar as I am claiming, then this semantic change accompanying the introduction of the copula would not be predicted, and this would be evidence against my hypothesis. However, an alternative view of ialah and adalah is possible, one which removes the conflict between my hypothesis and the data just discussed. As mentioned previously, both the words in question include a pragmatic particle in their form. One is clearly a grammaticalization of a left-dislocation construction: Xi, iai-lah Y, and the other quite possibly derives from a similar type of construction, as ada is an existential and presentative verb, very commonly initial in a clause: 59. Ada koran di meja exist newspaper LOC table 'There is a newspaper on the table.' It is therefore just as plausible to say that these two words are pragmatic elements used to clarify the structure of the clause, as it is to claim that they 223 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 are copulas. Then, the contrast in meaning of clauses with prefixed verbs depending on the presence of adalah or ialah would be predictable. The previous section has shown that prefixed verbs and their maximal projection can be nominalizations, and this would mean that a subject-predicate structure with such a phrase as predicate would be ambiguous between two readings, one with a nominal predicate and one with a verbal predicate. It is not hard then to imagine that adalah and ialah would come to be used to disambiguate, with their presence always indicating that a nominalised clause followed. This interpretation is supported by some facts about negation. Firstly, adalah and ialah are definitely not verbs as they cannot be preceded by the negator tidak. Any negation in a clause containing one of these words is a part of the predicate constituent following adalah or ialah: 60. Pernyataan ketua koperasi itu (*tidak) statement chairman co-operative that adalah tidak benar correct PRT NEG 'The statement by the chairman of the co-operative is not correct.' (IRG: 238) Secondly, when a nominalised clause functions as predicate, it can only be negated by the nominal negator bukan. When this word is present, there is no possible ambiguity and adalah and ialah 'rarely occur' (IRG: 308): 61. Kegemaran saya bukan memelihara binatang hobby 1SG animal NEG AS.take.care.of 'My hobby is not looking after animals' (IRG: 308) (Looking after animals is not my hobby) This can be compared with: 62. Kegemaran saya adalah tidak hobby 1SG PRT NEG 'My hobby is not looking after animals.' (Not looking after animals is my hobby) memelihara AS.take.care.of binatang animal These two pieces of evidence support the analysis of adalah and ialah as pragmatic particles which are used to clarify potentially ambiguous structures. On this interpretation, the meaning change that occurs when one of them precedes a prefixed verb is not counter-evidence to my hypothesis. Rather, the phenomenon is predicted. 5.3.5 Intransitive verbs: the prefixes ber- and terThe proposal being advanced here is that Indonesian has two types of clausal structure in which verbs can appear, and that a crucial diagnostic as to which type is being used is morphology: the prefixes meN- and di- derive potential predicates from verbal bases. Some one argument verbs in Indonesian are monomorphemic (e.g. pergi 'go', datang 'come'), but many are morphologically complex (e.g. ber-temu 'meet', ter-tidur 'fall asleep', and the verbs discussed in chapter 3). Section 3.3.3 already argued that the possibility of licensing an oblique argument by adjunction to the verb existed for one 224 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 type of intransitive verb, and the analysis presented here predicts that this possibility should be more widespread. The behaviour of some verbs prefixed with ber- and ter- supports this prediction. The prefix ber- derives intransitive verbs (IRG: 61-64, Verhaar 1984a), such as the following: 63. Bara cemburu berkobar lagi di hati Marisa fire jealous ber.flare.up again LOC liver Marisa 'The fire of jealousy flared up once more in Marisa's heart.' (SDM: 100) 64. Aku tidak mau bertengkar 1SG want ber.dispute NEG 'I don't want to quarrel.' (SDM: 71) But many verbs with this prefix can occur with a DP complement: 65. Arneta seorang janda berumur tiga Arneta one.CLASS divorcee ber.age three tahun year 'Arneta was a divorcee aged thirty four.' (SDM: 63) 66. Banyak siswa belajar bahasa many student ber.study language 'Many students study French.' (IRG: 266) puluh ten empat four Perancis France As would be predicted if such DPs are adjoined obliques, they cannot be separated from the verb: 67. *Arneta seorang janda berumur katanya tiga divorcee ber.age word.3 three Arneta one.CLASS puluh empat tahun ten four year (FOR: 'Arneta was a divorcee claiming to be aged thirty four.') 68. *Banyak siswa belajar dulu bahasa many student ber.study before language (FOR: 'Many students used to study French.') Perancis France And in some cases, an alternation between the prepositional and the adjoined realisations can be observed: 69. Semua orang ingin bertemu all person wish ber.meet 'Everyone wants to meet Lisa.' 70. Saya ingin bertemu guru saya 1SG wish ber.meet teacher 1SG 'I want to see my teacher.' (E&S: 565) 225 dengan with Lisa Lisa NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 I assume that in cases where such an alternation is not possible, this is due to the lack of a suitable preposition, as was argued in section 3.3.3. Similar data is discussed by Postal (1977), who suggests that ber- may be a marker of an antipassive clause type. Leaving aside the issue of whether this analysis is satisfactory, Postal also cites a personal communication from Dreyfuss to the effect that the post-verbal argument is restricted to generic and non-specific nominals, and that therefore this type of clause is an instance of noun incorporation14. Example 65, with a quantified nominal following the verb, and example 70, with a possessed nominal, suggest that this account is not correct (see also argumentation is section 2.3.1.2). The prefix ter- also generally derives intransitive verbs (IRG: 112ff) which often translate English passives, and carry an implication that the event was out of the control of the participant (Cartier 1978)15. A post-verbal nominal is possible with some ter- verbs (see also example 147 of chapter 4): 71. Dia sedang terlibat ter.involved 3SG PROG pegusaha wanita yang worker woman REL lebih tua more old 'He was having an affair with older.' (SDM: 59) affair dengan seorang affair with one.CLASS umurnya sepuluh tahun age.3 one.ten year a female worker who was ten years and an alternation between PP and DP occurs in some cases: 72. Obat itu terminum (oleh) anak saya medicine that ter.drink by child 1SG 'The medicine was accidentally drunk by my child.' (IRG: 115) The preposition is obligatory when a pronoun follows, a constraint which also applied to ke- -an verbs (see section 3.2.2.1), and the now familiar adjacency requirement also applies: 73. *Obat itu terminum dulu anak saya This data shows that the split between two types of clausal organization extends to the intransitive verbs of Indonesian, and confirms that there is an essential relation between the subject-predicate clause type and morphologically complex verbs. The possibility of an oblique adjoined to the verb should be limited to the case of morphologically complex (i.e. potentially predicative) verbs. Postal (1977: n44) suggests that the two analyses are not inconsistent; indeed he claims that antipassive in his sense is a prerequisite for incorporation. 14 Following the discussion of section 3.3.1, it is interesting to note that where ter- is prefixed to a verb root with a suffix, the suffix is almost always dropped and is always unacceptable for some speakers (IRG: 115-116). 15 226 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 5.3.6 Historical and comparative evidence The discussion of the preceding two sections of prefixed verbs heading nominalizations is related to some material presented in section 2.3.1.1. That section discussed the hypothesis put forward by Voskuil (1996) that the actor in a clause with a di-prefixed verb is syntactically equivalent to a possessor in a nominal projection, it has genitive case. The position of such actors and the fact that -nya can appear in this position are the main pieces of evidence in support of this position. Other material in chapter 2 made it clear that the various possibilities for the expression of the actor in clauses with di-prefixed verbs could not be treated together, which already suggests that Voskuil's analysis is not correct. The data presented in section 5.3.3 provides an additional argument against the proposal. If Voskuil is correct, we would expect that the enclitic could be used to represent the actor of a di-verb in a nominalisation. If such actors are in genitive case anyway, this would seem to be a natural prediction. But it is not correct. In the one construction where -nya must appear in a nominalisation, predicate nominalisation, it does not express the actor. Example 55, repeated here, does not have the meaning that some identifiable person or persons acted to close the roads; rather the nominalised clause has the reading of an agentless passive. 55. Lalu lintas menjadi macet karena ditutupnya traffic stuck because us.close.3 AS.become beberapa jalan several road 'Traffic became jammed because of the closure of several roads.' (IRG: 302) The function of -nya in such constructions is clearly non-referential as can be seen by comparing an intransitive clause with its nominalisation: 74. Kerusuhan terjadi di kampus campus riot occur LOC 'Riots occurred on the campus.' 75. Mereka prihatin atas terjadinya kerusuhan di 3PL concerned on occur.3 riot LOC kampus campus 'They were concerned about the occurrence of riots on the campus.' (IRG: 303) The pronoun does not represent a participant in the event here, and it does not do so in example 55 either. If anything can be analysed as a possessor in example 55, it is the subject of the original clause, the undergoer16. This is directly against the prediction which Voskuil's analysis would make. One function of -nya is to act as a ligature between the head noun and a possessor, especially where the structure might be ambiguous otherwise. For example, ibu Suparjo is ambiguous between 'Mrs Suparjo' and 'Suparjo's mother', while ibunya Suparjo has only the second meaning. 16 227 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 These considerations, as well as the fact that the projections of prefixed verbs are ambiguous between a nominal and a predicative reading, rule out any attempt to analyse Indonesian as having non-verbal predication as its predominant clause type. Nevertheless, it is valuable to consider the respects in which the current proposal resembles such analyses, in order to bring out the characteristic Austronesian traits which Indonesian possesses. This type of analysis has been proposed for all clauses in Tagalog and other Philippine languages (Dahl 1973: 117-118, Naylor 1995, Schachter & Otanes 1972: 62), and for non-actor subject clauses in Tagalog (Capell 1964, Lopez 1941), Atayal (Egerod 1966: 346), Toba Batak (van der Tuuk 1971) and Rukai (Li 1973: 202-211). The basic idea is that all the relevant clauses are equational, with a subject and a nominalised predicate. This can be illustrated using Tagalog examples. The following paradigm was given as examples 84-87 of chapter 2. It is repeated here with translations amended to indicate the equational analysis (the most natural English wording uses clefts which introduce an unwanted pragmatic effect unfortunately): 76. Bumili ang lalake ng isda ng pera sa man fish money OBL AF-buy NOM GEN GEN tindahan store 'The man is the one who bought fish with money in the store.' 77. Binili ng lalake ang isda ng pera sa man fish money OBL UF-buy GEN NOM GEN tindahan store 'The fish is what the man bought with money in the store.' 78. Binilhan ng lalake ng isda ng pera man fish money LF-buy GEN GEN GEN tindahan store 'The store is where the man bought fish with money.' 79. ang NOM Ipinambili ng lalake ng isda ang pera man fish money IF-buy GEN GEN NOM tindahan store 'The money is what the man bought fish with in the store.' sa OBL The morphology attached to the verb roots is assumed to have a particular character. It is derivational, and it orientates the word so that it may be used to refer to one of the participants involved in the event denoted by the base (Himmelmann 1991). Similar processes can be seen on a limited scale in English. The noun control is not oriented to any participant, and both arguments of the verb control can appear as modifiers in the noun's projection: Hoover's control of the FBI. But if the orientating nominalizer -er is used, the resulting noun refers to one of the participants and that argument cannot appear in the noun's projection: the/*Hoover's controller of the FBI. 228 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 Whilst I do not claim that prefixed verbs in general head nominalised constituents, the nature of the verb prefixes clearly has something in common with this view of Tagalog morphology. The prefixes are derivational, in that they prevent the root from projecting a regular verb phrase, and they are orientating, in that they specify the semantics of the argument of which they can be predicated. While I reject the view that prefixed verb clauses in Indonesian are equational, I leave open the possibility that they are attributive in nature. On this view, the prefix indicates that the subject of the predication has a certain role or attribute in regard to the event which is denoted. The sample of languages mentioned above as those for which equational sentence analyses have been proposed is biased towards the Formosan and Philippine languages which are generally assumed to have separated early in the Austronesian dispersal (Tryon 1995). This suggests that the clausal patterns which have suggested such analyses to various scholars were a feature of early stages of the history of the Austronesian language family. Starosta, Pawley & Reid (1982) argue that even if the Philippine-type systems we know today do not nominalise all verbs, the origin of such systems is in subject-predicate clauses in which the predicate was a nominalisation with an orientating affix. There is thus a considerable body of scholarly opinion which justifies treating the prefixed verb constructions of Indonesian as conservative in the context of the Austronesian language family. There is also good evidence that the bare verb constructions are innovative in the context of related languages. It is a characteristic of many Western Malayo-Polynesian languages that verbs can occur in two forms, a bare form and a form with a nasal segment prefixed to the root. These two forms are ordered differently with respect to the arguments of a transitive clause, the accessibility of the two arguments to syntactic processes varies with the verb form, and there may be a pragmatic difference in meaning between the two. These points are illustrated with examples from Balinese and Sasak (Ngenó-ngené variety): 80. Balinese Nyoman lempag tiang Nyoman hit 1SG 'I hit Nyoman' (Artawa & Blake 1997: ex2) 81. Tiang nglempag Nyoman Nyoman 1SG N.hit 'I hit Nyoman' (Artawa & Blake 1997: ex8) On the basis of various tests, Artawa and Blake show that the preverbal argument is the subject in each case, that is, the bare verb form is associated with undergoer subject. 82. Sasak Aku balé beli 1SG house buy 'It is a house that I buy.' (Austin 2000: ex3) 229 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 83. Aku mbeli balé 1SG house N.buy 'As for me, I buy a house.' (Austin 2000: ex4) The meaning difference here suggests that the bare verb form is more orientated towards the undergoer, and at least some syntactic processes show the same effect: only the patient can be questioned and can head a relative clause if the verb form is bare. Although it is not made evident in the glossing of the Tagalog examples above, it is also true in that language that bare verb forms are orientated towards the undergoer (Blake 1988, Himmelmann 1991)17. Starosta, Pawley & Reid (1982) claim that Proto-Austronesian had an ergative orientation, at least in part, and the sort of evidence just surveyed from daughter languages is compatible with such a claim. We would expect undergoer subject verb forms to be less-marked in ergative languages, and such a pattern might be retained from a proto-language18. In this context, the Indonesian bare verb clause types are unusual. The two types discussed previously in this study, bare verb actor subject clauses and emotion verb clauses, both resemble typical clauses of a language organised along nominative-accusative lines rather than the examples from other Austronesian languages examined above. These considerations, along with the tendency among grammarians to stigmatise these clause types, support the idea that they represent an innovative pattern in Indonesian. Section 2.2.1.4 discussed the historical origin of the prefix di-, and the evidence suggests that it is not a very old feature of the Malay subgroup. It is tempting to think that the introduction of this prefix and the introduction of actor subject bare verb clauses might be linked. If previously, the bare verb form occurred in undergoer subject clauses, the introduction of an alternative pattern might well provide the sort of pressure which might lead to innovative morphological marking on the older clause type. I stress that this is pure speculation: I have collected no data on the relative antiquity of the two phenomena in Malay19. 5.3.7 Summary The data and arguments presented here show that the idea that prefixed verbs in Indonesian have a good deal in common with non-verbal predicates in the language is plausible aside for reasons other than the contrasts which I seek to explain. However, all the issues discussed above relate to the status of the subject and the relation between it and the predicate. I am trying to account for contrasts in the behaviour of non-subject Aspect morphology is not glossed separately in the Tagalog examples for ease of reading. When this information is included, it becomes clear that the voice morpheme in example 66 is zero. 17 I do not intend to imply any claim that any of the languages discussed in the preceding paragraph are organised on ergative-absolutive lines. The issue has been debated for some years without any definite conclusions being reached. In addition to the works cited, see Maclachlan & Nakamura 1997, Manning 1996a, Payne 1982. 18 Voskuil (1996: 203-204) gives examples of bare verb actor subject clauses from classical Malay, but does not provide full details of the sources. 19 230 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 arguments. While it is a useful first step to be able to show that the constituents within which the non-subject arguments occur differ in status, this in itself is not an account of the differences in behaviour. I now turn to a discussion of the difficulty of expressing the intuition about the nature of prefixed verbs in LFG, and the closely related difficulty of giving a theoretically satisfying account of the different properties of the non-subject arguments of the two clause types. 5.4 Problems for an LFG account 5.4.1 Where does subj come from? Bresnan (2001a: 112) argues that the category S may be expanded as NP (or DP) and XP in a subject-predicate structure, that is, that S may be top node of a configurational structure, despite the exocentricity of the category. The phrase-structure schema in question, with functional annotation, is the following: 84. The origin of the SUBJ function in this schema raises some problems. The functional annotation makes it clear that the GF must be part of the lexical information associated with the head of XP: the ↑ = ↓ equation makes this clear. But the schema allows any category of head here. Is it plausible or desirable that, for example, all nouns have an item in their lexical argument list which maps to a SUBJ function in f-structure? Various solutions to this problem seem possible. Firstly, there is the possibility of a null copula analysis. This sort of account would assume that there is a lexical item in Indonesian which takes a SUBJ and a predicative phrase as its arguments, and which has no phonological content: 85. (PREDICATE) V < SUBJ, XP [+pred] > This alternative is allowed by the framework of LFG as it stands, although the introduction of null elements is against the spirit of the theory as I understand it. Another possibility is that the potentially predicative categories in Indonesian are systematically ambiguous. The LFG architecture does not really permit grammatical functions to be optional in lexical entries; optionality is a shorthand for multiple entries, with each one instantiating one option. We could assume that nouns, prepositions and adjectives in Indonesian all have two lexical entries, linked by redundancy rules: 231 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 86. N → N < SUBJ > One entry has no SUBJ function in its argument list, whilst the other does have this attribute. However, in line with the prior discussion, I assume that the grammatical function would be introduced directly by lexical specification. There is no contentful semantic role to be listed in the lexicon of these words, and therefore the GF cannot be the result of the operations of LMT. At most, the lexical forms with the SUBJ function have a semantic role, call it Attributee, available for the semantic interpretation of the SUBJ, although even this may not be a necessity. Note that these assumptions mean that every time one of those heads is introduced into a structure, two parallel f-structures must be created, one with a SUBJ function and one without. In every case, one of these structures will be ill-formed: where the item was not predicative, the fstructure with a SUBJ will be ruled out as incoherent (the GF would not be required by any predicate), and where the item was predicative, the f-structure without a SUBJ will be ruled out as incomplete (a GF required by a predicate would not be present). The approach is therefore computationally intractable, as the number of f-structures to be considered must increase exponentially. This is a considerable drawback. There is another disadvantage to this approach: it cannot capture the intuition behind the proposal introduced in section 5.3. I suggested there that the essential difference between prefixed verbs and unprefixed verbs was that they appeared in different types of structure. The significant difference between the two was, I proposed, that unprefixed verbs had a SUBJ function in their argument list, but that prefixed verbs did not. This approach allows both types to project the same sort of constituent in syntax (a VP) using the same phrase structure rules, while ensuring that only one of these constituents could result in a well-formed f-structure under the standard X-bar theory assumed. The prefixed verb would have no SUBJ function, the expansion of IP has the functional annotation (↑ SUBJ) = ↓ on its specifier node, by general principles, and the structure-function correspondence must therefore fail for such verbs. But if the SUBJ function is added to potential predicates by lexical redundancy rules, then this would no longer be the case, and nothing would prevent a prefixed verb from appearing in a standard SUBJ-VP structure. Some diacritic feature would need to be added to the prefixed verbs to ensure that this did not happen, and this in turn would require a separate set of phrase structure rules to account for the projections of prefixed verbs. Also, if a lexical redundancy rule similar to 86 applies to verbs, this must mean that the verb root has an argument which will map to SUBJ in its lexical entry (bare verbs have a SUBJ), the prefix must remove this when it makes the verb a potential predicate, and then the redundancy rule restores a SUBJ in a new lexical entry. This is messy and implausible. Bresnan (2001a: 276) proposes a solution to this problem as part of a discussion of predicative complements in English. While prepositions and nouns are only predicates of main clauses when supported by the copula, they can appear alone as predicates in embedded clauses with controlled subject functions: 232 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 87. Jogging keeps Susan [in a bad mood]. Bresnan proposes a lexical predication template 'which augments the lexical form for a preposition or nominal to one which has a subject of predication' (276): 88. 'in <(↑OBJ)>' ⇒ 'be-in <(↑SUBJ) (↑OBJ)>' This proposal differs from the redundancy rule approach discussed above in that in addition to altering the argument list of the lexical item, the template also alters the listed item. In the context of the discussion in which the proposal is made, it is not clear how this is to be understood. As a main clause predicate, a preposition such as in requires the copula, but in an XCOMP, the copula is impossible: 89. *Jogging keeps Susan [is/be in a bad mood]. Nonfinite verb forms can appear in such environments, therefore the impossibility of any form of the copula is surprising if it is indeed a part of the lexical item. If that is the intention, the template requires also a rule deleting be in non-finite environments. In any case, this approach will not transfer to Indonesian, unless it is taken to imply the zero copula analysis already discussed. In that case, it is simpler to assume that the zero copula licenses the SUBJ function, rather than including both steps. If on the other hand, the change in the lexical item is to be understood as semantic, then this proposal is identical to the redundancy rule discussed previously, as far as syntax is concerned. The case which Bresnan discusses as involving a clear semantic change, that of locative prepositions, also does not transfer to Indonesian where the purely locative preposition di can be used as a predicate: 90. Mereka di Jakarta sekarang Jakarta now 3PL LOC 'They are in Jakarta now.' (IRG: 237) The notion of a lexical predication template is therefore not helpful. There is additional evidence in Indonesian that a solution which assumes that the SUBJ is introduced by a lexical process, such as a redundancy rule, must be incorrect 20. Indonesian uses headless relative clauses freely. Such constituents have the same distribution as other DPs, they can fill argument positions: 91. Yang terbayang di depan matanya cuma Marisa visible front eye.3 only Marisa REL LOC 'What was visible before his eyes was Marisa only.' (SDM: 103) I am grateful to Peter Austin for pointing out the importance of the argument advanced in this paragraph. 20 233 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 92. Andi memilih yang besar Andy AS.choose big REL 'Andy chose the big one.' (IRG: 301). And, like other nominal constituents, headless relative clauses can be predicates: 93. Mobil saya yang biru blue car 1SG REL 'My car is the blue one.' (IRG: 301) This possibility is problematic for any account which assumes that the SUBJ function is added in the lexicon. There is no lexical head which projects the nominal constituent and which could have the SUBJ function associated with it by a lexical process. Three alternative analyses must be considered. Firstly, a null nominal head might be proposed for Indonesian, and the SUBJ could be associated with it in the relevant cases. This proposal is open to the same objection as the null copula analysis discussed previously. In addition, the null head would always and only co-occur with yang, as there is no independent motivation for null nominal heads in the language. Therefore, this analysis would be stipulative. The second possibility is that yang itself can acquire a SUBJ by lexical rule. This proposal is also stipulative because it would be necessary either to claim that yang is the only member of some lexical category, or that all other members of that category, perhaps complementizers, are barred from the redundancy rule which adds SUBJ to the lexical entry. It would not, however, be necessary to stipulate that yang could not appear as a predicate on its own. Nouns and adjectives have no obligatory arguments specified in their lexical entries, therefore they can be single-word predicates. Prepositions in Indonesian have obligatory complements and therefore can only be part of a phrasal predicate. A similar approach could be taken for yang: it has an obligatory clausal complement. This proposal would also add to the computational problem discussed before. If yang has two possible lexical entries, and every noun does also, any nominal constituent including a relative clause would have four possible f-structures associated with it. The final alternative is that yang always introduces a predicate, and that the head noun which is modified by the relative clause is always its SUBJ. Such an analysis is semantically awkward, in that the head noun always has some function in the relative clause, and syntactically improbable, in that it would imply a claim that any DP including a relative clause was a categoryshifted clause. None of these three analyses is plausible, therefore the use of headless relative clauses as predicates is strong evidence against any proposal which allows potential predicates to acquire a SUBJ function by some lexical process. A solution to the problem under consideration would be to have the phrase structure rule which allows DP, XP as an expansion of S to be associated with some f-structure mechanism which introduces the SUBJ function. Such a proposal would capture the intuition that the non-verbal categories are predicative in only this syntactic environment. However, this solution conflicts with a basic principle assumed in LMT, the subject condition: 234 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 Every lexical form must have a subject. (Bresnan & Kanerva 1989: 28) or: Every predicator must have a subject. (Bresnan 2001a: 311) This principle apparently rules out any analysis in which the introduced constructionally. SUBJ function is However, the subject condition is qualified in both the sources cited above. Bresnan & Kanerva (1989) note that there are languages, such as Acehnese (Durie 1985), for which no clear case can be made for analysing a subject grammatical function. Additionally, Bresnan (2001a) specifies that the condition applies only to the a-structures of predicators taking core arguments, that is SUBJ, OBJ and OBJθ. Even on this basis, the condition does not apply in all cases. For example, Bresnan (2001a: 275-280) analyses locative PPs in English as oblique arguments. The complement of the preposition is analysed as an OBJ, a core argument function, and the subject condition might be expected to apply to the prepositional predicator. But it is explicitly stated (p277) that such structures lack a subject of predication. These considerations suggest that the statements of the subject condition quoted above are not adequate, and that this condition should be considerably more restricted in its scope21. Therefore I will assume that, in restricted circumstances, grammatical functions can be introduced by non-lexical means. The concrete details of my proposal would be as follows22. The phrase structure rule allowing subject-predicate structures also introduces a PRED value ‘is-predicated-by’ taking a SUBJ and a PREDICATE as its arguments. This item is not associated with any c-structure position: 94. S → [PRED ‘is-predicated-by’ < SUBJ PREDICATE > ] where predicate = XP [PREDICATIVE +] DP XP In the case of a simple nominal predicate, this rule allows structures of the following type: Joan Bresnan (p.c.) suggests that “the subject condition must be restricted to a subset of verbal predicators”. 21 I am grateful to Joan Bresnan for suggesting improvements to the formulation of this proposal. 22 235 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 95. 96. In the case of lexical items other than verbs, the feature [PREDICATIVE +] is assigned to all lexical heads by a redundancy rule. The feature is relevant for subject-predicate structures and for the distribution of enclitic pronouns and its presence has no other effects. Verbs only receive this feature as a result of the morphological operation of prefixation, and therefore bare verbs cannot fill the place of the XP in rule 94. 5.4.2 Status of the non-subject argument There are two issues to consider in relation to the status of the nonsubject argument of a meN-prefixed verb: what grammatical function is assigned to such arguments, if any; and how is that function assigned? I will propose two possible answers to the first question, both of these will lead naturally to the discussion of the second issue. One possible answer to the question of what GF is assigned to the nonsubject argument of a meN-prefixed verb is to say that it is an OBJ, and that any differences between such arguments and the non-subject arguments of bare verbs are due to the fact that one is inside a predicative constituent and the other is not. Such an analysis implies a stipulation that arguments cannot be extracted from within predicative constituents. This stipulation is empirically accurate, but is not derivable from other properties of the construction discussed here. The advantage of this approach is its intuitive appeal: the non-subject arguments of prefixed verbs have a good deal in common with the non-subject arguments of bare verbs. They are certainly terms, and non-subject terms are generally OBJs. The disadvantage is that this account requires the stipulation as to the impossibility of extracting from predicative constituents. This is clearly ad hoc: it is not possible to extract from other potentially predicative constituents when they appear in either predicative or non-predicative environments, nor is it possible to extract from nominalizations of clauses: 236 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 97. *buku yang membaca menambahkan pengetahuan book knowledge REL AS.read AS.increase umum common (FOR: 'books of which the reading increases general knowledge') Thus, the stipulation would only be inserted to ensure the desired result in the one case where extraction would be predicted to be possible. The alternative approach is to assume that the non-subject argument of a prefixed verb is not an OBJ. Then the constraint on forming relative clauses can be stated in terms of a natural class of arguments defined by LMT: only [-r] arguments (SUBJ and OBJ) can head gapped relative clauses. This is preferable to the stipulation discussed in the previous paragraph, but leaves the questions about the status of the non-subject argument still unanswered. Evidence has previously been given that this argument is a term: it can float a quantifier, and in the special case of the ergative undergoer subject clauses, it can antecede a reflexive which is also a term. This means that the only possibilities other than OBJ are OBJθ, or some GF not currently recognised in LFG. The reasons discussed in section 4.4.1 for not introducing another GF are compelling, but so were the reasons discussed there for not allowing OBJθ as the first non-subject argument of a verb. These reasons are reinforced in this case by the fact that prefixed verbs with two objects occur in Indonesian. If it is strange to analyse the single non-subject argument of a verb as an OBJθ, then it is even stranger to analyse both non-subject arguments of a ditransitive verb as having that GF. It is interesting to note that the same point has been reached in chapter 4 and here, arguing from opposite assumptions: if the prefixed verbs are true transitive verbs, then the non-subject arguments of the bare verbs must have a GF other than OBJ with OBJθ the only available option; and if the prefixed verbs are not true transitive verbs, then the same must be true of their non-subject arguments. Again, this suggests that the LFG framework is too restricted to accurately describe the data in this case. One possible solution is to appeal to the distinction between terms (direct arguments) and non-terms (obliques and adjuncts). The version of a(rgument)-structure proposed by Manning (1996a,b) Arka and Manning (to appear) and Wechsler and Arka (1998) incorporates information as to the term or non-term status of arguments. Section 3.3.2.3 has presented evidence that this version of a-structure gives a better account of some Indonesian data, and if we accept this version of a-structure, then the notion term is a part of the theory. Bresnan (2001a: 96) recognises a core / non-core distinction, but it is not clear what theoretical consequences this distinction has. The notion is independently required in any case. Many languages distinguish terms from non-terms in their grammar, by restricting certain syntactic processes to apply only to one class. Where terms includes OBJθ, the class of terms cannot be stated as a natural class using the feature specifications of LMT. SUBJ and OBJ share the feature [-r], OBJ and OBJθ share the feature [+o], but SUBJ and OBJθ do not share any feature. Some notion such as term is necessary in these cases in order to state syntactic generalisations. We might then allow, in special circumstances, an argument to be licensed in an f-structure purely by virtue of 237 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 being a term. Such arguments would not derive their GF from the operation of LMT, and would therefore have no feature specification. If prefixed verbs have non-subject arguments of this type, then the constraint on gapped relativization suggested previously would give the desired results. However, this is almost as much a stipulation as the alternative solution to the problem discussed previously, with the proviso that the term / non-term distinction is a fact of Indonesian grammar: terms can float quantifiers and non-terms cannot. I will advance only one more speculation on this topic. The previous paragraph already suggested that it might be possible that LMT does not operate on the SUBJ argument of prefixed verbs, and by implication, of other potential predicates. It might be possible then that LMT does not apply to potential predicates at all. This idea has some plausibility if we consider the other types of head which can be predicates. Nouns can have a possessor argument, but this is a property of the category N, and is not mediated by LMT. Adjectives (at least in Indonesian) do not license a direct argument (see section 4.1.3) and an oblique is the only possibility for a non-subject argument in that case. Prepositions are perhaps more problematic: an argument introduced by a preposition is an oblique, but what GF is assigned to the complement of the preposition: is it also an OBLθ, or is it an OBJ as traditional terminology might suggest? Complement of prepositions cannot be athematic, therefore I assume that OBLθ is the correct answer. But in either case, it is plausible that the GF is not assigned via LMT. However, it is not at all clear that similar arguments can be made in regard to prefixed verbs. Apart from anything else, there are ditransitive verbs to be accounted for, and although this approach may give an answer to the question of how nonsubject arguments of predicates receive a GF, it cannot answer the question of what GF the non-subject argument of a prefixed verb actually has. I am forced to conclude that this question is not answerable in a satisfying way in LFG as currently formulated. 5.5 Conclusions I have argued in this chapter that the contrasting properties of nonsubject arguments of prefixed and bare verbs suggest a divide between two types of clause structure in Indonesian. There is evidence to support the idea that the type of structure that prefixed verbs appear in has much in common with non-verbal predications, and this idea also fits well into an historical and comparative view of Austronesian syntax. The type of clause in which bare verbs appear has more in common with the subject-VP structure familiar to linguists trained in English-speaking countries, and this is an innovative pattern in the context of the Austronesian family. I have also argued that there is no straightforward way to express this idea using the available theoretical vocabulary of LFG. To the extent that the intuition underlying my proposal regarding prefixed verbs is correct, this conclusion exposes inadequacies in the theory. In terms of the objectives of this study as a whole, these conclusions mean that it has not been possible to give a complete account of non-subject arguments in Indonesian. However, the investigation has produced interesting results: 238 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 - morphological realisation of arguments can allow the possibility of ergative clauses in a language whose basic alignment is not ergative-absolutive; - Indonesian allows the OBLθ grammatical function to be licensed by adjunction to a morphologically complex verb where the argument is assigned a semantic role by the verb. In one case, this strategy was the only one available as no preposition exists in the language which is semantically compatible with the relevant semantic role, patient; - Indonesian also requires the OBLθ grammatical function in one clause type to be realised with a preposition although the semantic role assigned to the argument comes direct from the predicate. The preposition does not make a semantic contribution; - there is an example of predicate composition in the language where the composed argument structure is the result of purely semantic considerations; - there is evidence that the language has (at least) two types of clausal organization side-by-side. - the mapping of arguments to GFs is satisfactorily accounted for by LMT in only one of the two types of clause, the innovative one. LMT is of limited use in analysing the conservative clause structure, and this finding is in line with a number of recent studies of Austronesian languages (Arka 1998, Foley 1998, Sells 1998, Wechsler & Arka 1998). Indonesian has strategies for the licensing of arguments outside of the scheme introduced at the beginning of this work. Two of these seem to be of particular importance, and these two are linked in an important fashion. Firstly, terms can be expressed morphologically in Indonesian and secondly oblique arguments can be licensed by adjunction to the verb which assigns them a semantic role. In both cases, adjacency to the head which governs the argument and the morphological composition of that head are essential features of the licensing mechanism. Both of these processes are possible only in clauses which use the conservative type of structure identified in chapter 5. This suggests that better understanding of that type of clausal structure and of the ways in which arguments are licensed in it will depend in part on investigation of the historical development of Indonesian (or Malay) syntax. Two other questions for future research arise from the results reported here: - to what extent does the split in clausal organization extend to intransitive verbs? The discussion of chapter 3 and of section 5.3.5 has brought forward some suggestive data, but a more detailed analysis of the problem is needed. - what is the significance of the particular properties attributed to functional categories in Indonesian in this work, and do they correlate with other typological characteristics? 239 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 Appendix: Indonesian and the accessibility hierarchy Keenan & Comrie (1977; afterwards K&C) propose some universal restrictions on the formation of relative clauses. One of the languages in their sample is Malay, although the nature of the data used is not entirely clear. K&C acknowledge the assistance of Chung with Indonesian data, they cite McDonald & Dardjowidjojo (1967), which is a reference grammar of Indonesian, but they also state in a footnote (n2) that their two informants were from Malaya. Whatever the source and status of their data, K&C's claims do not tally with my understanding of Indonesian grammar, and the actual facts may constitute a counter-example to their second universal Any R(elative)C(lause)-forming strategy must apply to a continuous segment of the A(ccessibility)H(ierarchy). K&C report that Malay has two relativization strategies, one of these is, in their terms, [-case] which is the gapped strategy. The second is [+case] in K&C's terms, because a preposition can precede the head noun. The example which they give of this strategy is distinct from either of the strategies described above: 98. perempuan kepada siapa Ali beri ubi kentang woman to who Ali give potato 'the woman to whom Ali gave the potato' (K&C: ex15) itu that K&C acknowledge that this example was elicited under pressure, and it is judged strange by native speakers. The summary table given by K&C shows this strategy to be possible for indirect objects, obliques and genitives. The primary strategy, gapping, is shown as possible for subjects and some objects. However, the example which is given to show object relativization is ungrammatical according to most sources and according to native speakers: 99. ?*Ali bunuh ayam yang Aminah sedang memakan Aminah PROG Ali kill chicken REL AS.eat 'Ali killed the chicken that Aminah is eating.' (K&C: ex13) Unlike the example used by Chung (1976a) in making a similar point, there is no question here as to the type of clause which is embedded: it is clearly a meN-V clause (see discussion in section 1.1.2). Presenting the facts of interest in K&C's study as seen in the data given in this study reveals a problem. If we equate the non-subject argument of all transitive structures with K&C's Direct Object, then it is true that the primary strategy is available for all subjects and some objects, the same result that K&C report, if for rather different reasons. And if we take the resumptive pronoun strategy as the secondary strategy, ignoring the possibility shown in K&C's example quoted above (example 96), this strategy is available for some objects, some obliques and genitives. A comparison of the two sets of results looks like this23: 23 I omit the column for object of comparison, which is not relevant. 240 NON-SUBJECT ARGUMENTS IN INDONESIAN − CHAPTER 5 100. K&C Strategy 1 Strategy 2 Musgrave Strategy 1 Strategy 2 Subj + + - Relativizable Positions DObj IObj Obl +/+ + +/+/- ? +/- Gen + + It is not immediately clear how to include K&C's indirect object in the present framework. Although the Accessibility Hierarchy appears to generalise over syntactic positions, K&C treat indirect object as a semantic entity. The example from Malay given previously (example 96) apparently is the basis for the claim that indirect objects are relativizable in Malay. But the recipient in example 96, and in the simple clause related to the embedded clause in that example, is formally an oblique: it is introduced by a preposition. In addition, K&C do not seem to recognise second objects as relevant to their investigation, and it is hard to see how they would integrate that grammatical function into their framework. Ditransitive verbs in asymmetrical object languages (Bresnan & Moshi 1990) have the recipient (K&C's indirect object) as their object, with the theme argument as second object. If we reconstruct the accessibility hierarchy as a syntactic phenomenon, and use the meta-language of LFG for this, the following is the result for Indonesian: 101. Relativizable Positions Strategy 1 Strategy 2 SUBJ OBJ + - +/+/- OBJθ - OBLθ + POSS + If this is a licit reinterpretation of K&C's hierarchy, then Indonesian is a counter-example to one of the universals that they propose: the secondary strategy does not apply to a continuous portion of the hierarchy. For LFG, this result is not surprising. As discussed in section 1.3.3, LMT predicts that OBJθ will be the most marked GF. OBJ and OBLθ would be adjacent by the predictions of the theory, according more closely with the Indonesian facts. The relative ranking of OBJθ and POSS would still be an issue, but I will assume that a plausible argument can be made that possessors are more common in the languages of the world than second objects. Therefore, on theory-internal grounds, an LFG account gives a hierarchy which more accurately reflects the Indonesian facts as well as avoiding the problem in K&C's account of mixing syntactic and semantic categories in formulating their hierarchy. 241 REFERENCES Sources of Examples E&S - John M. Echols & Hassan Shadily (1994) Kamus Indonesia-Inggris Jakarta: Penerbit PT Gramedia (3rd edition revised and edited by John U. Wolff and James T. Collins assisted by Hassan Shadily) PYD - Fredy S (1991) Perasaan yang ditinggalkan Jakarta: Gultom Agency SDM - Mira W (1995) Sekelam dendam Marisa Jakarta: Penerbit PT Gramedia SDPGS - Agatha Christie, tr. Alex Tri Kantjono W (1986) Satu, Dua, Pasang Gesper Sepatunya Jakarta: Penerbit PT Gramedia References Abney, Stephen P. (1987) The English Noun Phrase in its Sentential Aspect. PhD Dissertation, MIT. Adams, K. & A. Manaster-Ramer (1988) Some questions of Topic/Focus choice in Tagalog Oceanic Linguistics 27: 79-101 Adelaar, K. Alexander (1992a) The relevance of Salako for Proto-Malayic and for Old Malay epigraphy. Bijdragen tot de Taal-, Land- en Volkenkunde 148:381-408 Adelaar, K. Alexander (1992b) Proto Malayic: The reconstruction of its phonology and part of its lexicon and morphology (Pacific Linguistics C-11) Canberra: Research School of Pacific and Asian Studies Aissen, Judith (1999) Markedness and subject choice in Optimality Theory. Natural Language and Linguistic Theory 19:673-711 Alsagoff, Lubna (1991) The Notions of Subject and Topic in Malay. Proceedings of the Berkeley Linguistics Society 16: 21-75 Alsagoff, Lubna (1992) Topic in Malay: the other subject PhD Dissertation, Stanford University Alsina, Alex (1997) A Theory of Complex Predicates: Evidence from causatives in Bantu and Romance. In Alex Alsina, Joan Bresnan & Peter Sells (ed) Complex Predicates Stanford CA: CSLI Publications Anderson, Edmund A. (1983) The meaning of variation in Indonesian. Linguistic Studies of Indonesian and Other Languages in Indonesia (NUSA 15) pp1-26 Anderson, Mona (1983) Prenominal genitive NPs. The Linguistic Review 3: 1– 24 Andrew, Avery D. (1984) Lexical insertion and the elsewhere principle in LFG. Unpublished manuscript, ANU. Andrews, Avery D. (1990) Unification and morphological blocking. Natural Language and Linguistic Theory 8: 507–557 Andrews, Avery D. & Christopher D. Manning (1999) Complex Predicates and Information Spreading in LFG Stanford CA: CSLI Publications Aristar, Anthony R. (1997) Marking and hierarchy types and the grammaticalization of case-markers Studies in Language 21: 313-368 242 REFERENCES Arka, I Wayan (1993) Morpholexical aspects of the -kan causative in Indonesian MPhil thesis, University of Sydney Arka, I Wayan (1998) From morphosyntax to pragmatics in Balinese PhD thesis, University of Sydney Arka, I Wayan & Christopher D. Manning (to appear) Voice and grammatical relations in Indonesian: A new perspective. In Peter K. Austin and Simon Musgrave (ed) Voice and Grammatical Functions in Austronesian Arka, I Wayan & Jane Simpson (1998) Control and complex arguments in Balinese. In Miriam Butt & Tracy Holloway King (eds) Proceedings of the LFG98 Conference CSLI Publications. Available at: http://cslipublications.stanford.edu/LFG/3/arka-simpson.html Artawa, Ketut (1999) Ergativity and Balinese Syntax (3 parts) NUSA 42 Artawa, Ketut & Barry J. Blake (1997) Patient primacy in Balinese. Studies in Language 21: 483–508 Artawa, Ketut, Put Artini & Barry J. Blake (1997) Balinese Grammar and Discourse. Unpublished MS, Udayana University, STKIP, Singaraja and La Trobe University. Austin, Peter (1997a) Causatives and applicatives in Australian Aboriginal Languages, in Kazuto Matsumura and Tooru Hayasi (eds) Dative and related phenomena, pp165-225. Tokyo: Hitsuji Shobo. Austin, Peter (1997b) Ergativity, clitics and grammatical relations in Sasak. In M.Butt & T.Holloway (ed) Proceedings of the First Lexical Functional Grammar Conference, Rank Xerox Research Centre, Grenoble Palo Alto CA: Stanford University Austin, Peter K. (2000) Verbs, voice and valence in Sasak. In Peter K. Austin (ed) Working Papers in Sasak, vol.2 pp5-24, Melbourne: Lombok and Sumbawa Research Project, The University of Melbourne Austin, Peter K. (to appear) Lexical-Functional Grammar. In International Encyclopedia of Social and Behavioural Sciences (Vol.3, part 9 article 20) London: Elsevier Scientific Baker, Mark (1988) Incorporation Chicago: University of Chicago Press Belletti, Adriana & Luigi Rizzi (1988) Psych verbs and Τ-theory. Natural Language and Linguistic Theory 6:291-352 Benjamins, Geoffrey (1997) Affixes, Austronesian and Iconicity in Malay. Department of Sociology Working Papers No.132 Singapore: National University of Singapore Blake, Barry J. (1977) Case-marking in Australian Languages Canberra: Australian Institute of Aboriginal Studies Blake, Barry J. (1988) Tagalog and the Manila-Mt Isa axis LaTrobe Working Papers in Linguistics 1: 77-90 Blake, Barry J. (1990) Relational Grammar London/New York: Routledge Blake, Barry J. (1994) Case Cambrdige UK: Cambridge University Press Blake, Frank R. (1906) Expression of case by the verb in Tagalog. Journal of the American Oriental Society 27: 183–189 243 REFERENCES Blevins, J.P. (1995) Syncretism and paradigmatic opposition Linguistics and Philosohy 18: 113-152 Bloomfield, Leonard (1917) Tagalog Texts with Grammatical Analysis Urbana: The University of Illinois (University of Illinois Studies in Language and Literature Vol. III, part 2-4) Blume, Kerstin (1998) A contrastive analysis of interaction verbs with dative complements Linguistics 36: 253-280 Bresnan, Joan W. (1973) Syntax of the comparative clause construction in English Linguistic Inquiry 4:275-343 Bresnan, Joan W. (1982) Control and complementation. Linguistic Inquiry 13:343-434 Bresnan, Joan W. (2001a) Lexical-Functional Syntax Malden MA/Oxford UK: Blackwells Publishers Bresnan, Joan W. (2001b) Explaining morphosyntactic competition. In Mark Baltin & Chris Collins (eds) Handbook of Contemporary Syntactic Theory pp1144, Oxford UK: Blackwells Publishers Bresnan, Joan W. & Jonni Kanerva (1989) Locative inversion in Chichewa: A case study of factorization in grammar. Linguistic Inquiry 20: 1-50 Bresnan, Joan W. & Sam A. Mchombo (1987) Topic, pronoun and agreement in Chichewa. Language 63: 741-782 Bresnan, Joan W. & Lioba Moshi (1990) Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21: 147-185 Burzio, Luigi (1986) Italian Syntax: a Government and Binding approach Dordrecht: Reidel Capell, Arthur (1964) Verbal systems in Philippine languages Philippine Journal of Sciences 93: 231-248 Cardinaletti, Anna & Michal Starke (1996) Deficient pronouns: A view from Germanic. In Höskuldur Thráinsson, Samuel David Epstein & Steve Peter (eds) Studies in Comparative Germanic Syntax, Vol. 2 pp21-65. Dordrecht: Kluwer (Studies in Natural Language and Linguistic Theory 38) Carter, Richard (1976) Some linking regularities. In Beth Levin & Carol Tenny (ed) On Linking: Papers by Richard Carter Cambridge MA: Center for Cognitive Science, MIT (Lexicon Project Working Papers No.25) Cartier, Alice (1978) On ke-verb sentences in Indonesian. In S.A.Wurm & Lois Carrington (ed) Second International Conference on Austronesian Linguistics Proceedings, Fascicle 1 Canberra: Pacific Linguistics (Series C-61) Cartier, Alice (1979) De-voiced Transitive Verb Sentences in Indonesian. In Frans Plank (ed.) Ergativity: Towards a theory of grammatical relations pp161-183 London: Academic Press Cena, Resty (1995) Surviving without relations. MS, Edmonton, Alberta. Chafe, Wallace L. (1970) Meaning and the Structure of Language Chicago: University of Chicago Press Chomsky, Noam (1957) Syntactic Structures The Hague: Mouton Chomsky, Noam (1981) Lectures on Government and Binding Dordrecht: Foris Publications 244 REFERENCES Chomsky, Noam (1991) Some notes on economy of derivation and representation. In R. Friedin (ed) Principles and Parameters in Comparative Syntax Cambridge MA: MIT Press (also in Chomsky The Minimalist Program MIT Press [1995]) Chomsky, Noam & Howard Lasnik (1995) The Theory of Principles and Parameters. In N.Chomsky The Minimalist Program pp13-127, Cambridge MA: MIT Press Chung, Sandra (1976a) An Object-Creating Rule in Bahasa Indonesia. Linguistic Inquiry 7:41-87 Chung, Sandra (1976b) On the subject of two passives in Indonesian. In Charles Li (ed) Subject and Topic pp57-98, New York: Academic Press Chung, Sandra (1978a) Stem sentences in Indonesian. S.Wurm & L.Carrington (ed.) Second International Conference on Austronesian Linguistics: Proceedings Fascicle 1 (Pacific Linguistics C-61) pp335–365 Canberra: Research School of Pacific and Asian Studies Chung, Sandra (1978b) Case Marking and Grammatical Relations in Polynesian Austin: University of Texas Press Clynes, Adrian (1995) Topics in the Phonology and Morphosyntax of Balinese PhD dissertation, Australian National University Cole, Peter & Gabriella Hermon (2000) Word order and binding in Toba Batak. Paper presented at AFLA 7, Amsterdam. Cumming, Susanna (1991) Functional Change: The Case of Malay Constituent Order Berlin: Mouton de Gruyter Dahl, Otto Christian (1973) Proto-Austronesian Lund: Studentlitteratur (Scandanavian Institute of Asian Studies Monograph 15) Dardjowidjojo, Soenjono (1978) The semantic structure of the adversative ke-an verbs in Indonesian. In S.Udin (ed) Spectrum: Essays presented to Sutan Takdir Alisjahbana on his seventieth birthday pp107-124, Jakarta: Dian Rakyat Davies, William & Carol Rosen (1988) Unions as multi-predicate clauses Language 64: 52-88 De Casparis, J. G. (1956) Selected inscriptions from the 7th to the 9th century A.D. Bandung : Masa Baru (Prasasti Indonesia ; no. 2) De Guzman, Videa P. (1988) Ergative analysis for Philippine languages: an analysis. In R. McGinn (ed) Studies in Austronesian Linguistics Athens OH: Center for International Studies (Ohio University Monographs in International Studies, Southeast Asia Series 76) De Guzman, Videa P. (1997) Some remarks on the Grammatical Function of the 'Non-Subject' Agent in Tagalog. Paper presented at the 8th International Conference on Austronesian Linguistics, Academia Sinica, Taipei. Di Sciullo, Anne-Marie & Edwin Williams (1987) On the Definition of Word Cambridge MA: MIT Press Dixon, R.M.W. (1977) Where have all the adjectives gone? Studies in Languge 1:1-80 Dixon, R.M.W. (1979) Ergativity Language 55: 59-138 Dixon, R.M.W. (1994) Ergativity Cambridge UK: Cambridge University Press 245 REFERENCES Donohue, Mark (1999) Syntactic roles vs. semantic roles: external possession in Tukang Besi. In Doris L. Payne & Immanuel Barshi (ed) External Possession pp373-401, Amsterdam/Philadelphia: John Benjamins Dowty, David R. (1991) Thematic proto-roles and argument selection. Language 67: 547-619 Dryer, Matthew (1986) Primary objects and secondary objects Language 62: 808-845 Durie, Mark (1985) A Grammar of Acehnese Foris: Dordrecht Echols, John M. and Hassan Shadily (1961/1989) Kamus Indonesia-Inggris Jakarta: Penerbit PT Gramedia (3rd edition revised by John U. Wolff and James T. Collins in association with Hassan Shadily) Egerod, Søren (1966) Word order and word classes in Atayal Language 42: 346-369 Falk, Yehuda N. (to appear) Lexical-Functional Grammar: An introduction to parallel, constraint-based syntax Stanford CA: CSLI Publications Fillmore, Charles J. (1968) The case for case. In E.Bach & R.T.Harms (ed) Universals in Linguistic Theory pp1-88, London: Holt, Rinehart & Winston Foley, William A. (1991) The Yimas Language of New Guinea Stanford CA: Stanford University Press Foley, William A. (1998) Symmetrical voice systems and precategoriality in Philippine languages. Paper presented at the 3rd Lexical-Functional Grammar Conference, Brisbane. Available from http://www.sultry.arts.usyd.edu.au/LFG98/austro/workshop.htm. Foley, William A. and Robert D. Van Valin Jnr (1984) Functional syntax and universal grammar Cambridge UK: Cambridge University Press Givón, Talmy (1983) Topic continuity in discourse: An Introduction. In T.Givón (ed) Topic Continuity in Discourse: A Quantitative Cross-Language Study pp541, Amsterdam: John Benjamins Gonda, J. (1949) Prolegomena tot een theorie der woordsoorten in Indonesische talen. Bijdragen tot de Taal-, Land- en Volkenkunde 105: 275331 Grimes, Barbara F. (ed. 1996) Ethnologue (13th edition) Dallas: Summer Institute of Linguistics Grimshaw, Jane (1990) Argument Structure Cambridge MA: MIT Press Grimshaw, Jane (1997) Projection, heads and optimality Linguistic Inquiry 28:373-422 Gruber, J.S. (1965) Studies in Lexical Relations PhD Thesis, MIT Guilfoyle, Eithne, Henrietta Hung & Lisa Travis (1992) SPEC of IP and SPEC of VP: Two subjects in Austronesian languages. Natural Language and Linguistic Theory 10: 375–414 Haegeman, Liliane (1991) Introduction to Government & Binding Theory Oxford: Blackwell Publishers Hall, Barbara (1965) Subject and object in modern English PhD dissertation, MIT Harris, Z.S. (1965) Transformational theory Language 41: 363-401 246 REFERENCES Hengeveld, Kees (1992) Non-verbal Predication Berlin/New York: Mouton de Gruyter Hilman & Boim (1992) Lupus. Ih, Syereem! Jakarta: Gramedia Pustaka Utama Himmelmann, Nikolaus P. (1991) The Philippine Challenge to Universal Grammar Köln: Institut für Sprachwissenschaft (Arbeitspapier No.15, Neue Folge) Himmelmann, Nikolaus P. (1996) Person marking and grammatical relations in Sulawesi. In H. Steinhauer (ed) Papers in Austronesian Linguistics No. 3 pp115-136, Canberra: ANU (Pacific Linguistics A-84). Hiorth, Finngeir (1976) Active/passive and non-substantival ke- -an forms in Bahasa Indonesia Review of Indonesian and Malayan Affairs 8:62-88 Hopper, Paul (1979) Some discourse origins of ergativity, Hawaii Working Papers in Linguistics II:137–153. Hopper, Paul (1988) How ergative is Malay? In R.McGinn (ed) Studies in Austronesian Linguistics pp441-454, Athens OH: Ohio University Press Hopper, Paul & Sandra Thompson (1980) Transitivity in grammar and discourse Language 56: 251-299 Iskandar, N.S. (1946) Djangir Bali Jakarta: Balai Pustaka Iskandar, N.S. (1952) Udjian Masa Jakarta / Groningen: J.B.Wolters Jackendoff, Ray (1987) The status of thematic relations in linguistic theory. Linguistic Inquiry 18:369-411 Jackendoff, Ray (1990) Semantic Structures Cambridge MA: MIT Press Jaeggli, Osvaldo A. (1986) Passive Linguistic Inquiry 17: 587-622 Jelinek, Eloise (1984) Empty categories, case, and configurationality Natural Language and Linguistic Theory 2: 39-76 Junus, Umar (1971) ke- -an construction in Indonesian Linguistics 17: 24-44 Kachru, Yamuna (1990) Experiencer and Other Oblique Subjects in Hindi. In M.K.Verma and K.P.Mohanan (ed)Experiencer subjects in South Asian languages pp59-75, Stanford CA: CSLI Publications Kana, Marit (1986) Grammatical Relations in Bahasa Indonesia Cornell University: PhD dissertation Kaplan, Ronald M. & Joan W. Bresnan (1982) Lexical-Functional Grammar: A formal system for grammatical representation. In Joan w. Bresnan (ed) The Mental Representation of Grammatical Relations pp173-281, Cambridge MA: MIT Press Kaswanti Purwo, Bambang (1984) The categorial system in contemporary Indoensian: pronouns. In John M. Verhaar (ed) Towards a description of contemporary Indonesaian: preliminary studies Part 2 (NUSA 19) pp55-74 Kaswanti Purwo, Bambang (1988) Voice in Indonesian: A discourse study. In M.Shibatani (ed) Passive and Voice pp195-241, Amsterdam: John Benjamins Keenan, Edward L. (1976) Towards a universal definition of 'subject'. In Charles Li (ed) Subject and Topic pp303-333, New York: Academic Press Keenan, Edward L. (2000) Morphology is structure: a Malagasy test case. In I. Paul, V. Phillips & L. Travis (ed) Formal Issues in Austronesian Linguistics pp27-48, Dordrecht: Kluwer 247 REFERENCES Keenan, Edward L. & Bernard Comrie (1977) Noun Phrase accessibility and Universal Grammar Linguistic Inquiry 8: 63-99 Kiparsky, Paul (1982) From Cyclical Phonology to Lexical Phonology. In Harry van der Hulst and Norval Smith (ed) The Structure of Phonological Representations (Part 1) pp131-175, Dordrecht: Foris Publications Kroeger, Paul R. (1993) Phrase Structure and Grammatical Relations in Tagalog Stanford CA: CSLI Publications Kroeger, Paul R. (1998) Nouns and verbs in Tagalog: A reply to Foley. Paper presented at the 3rd Lexical-Functional Grammar Conference, Brisbane. Available from http://www.sultry.arts.usyd.edu.au/LFG98/austro/workshop.htm. Kroon, Yosep (1998) The isiq construction and its grammatical relations. In P.K. Austin (ed) Working Papers in Sasak Vol.1 pp105-118, Melbourne: Lombok and Sumbawa Research Project, The University of Melbourne Larson, Richard K. (1988) On the Double Object construction. Linguistic Inquiry 19: 335-391 Legendre, Géraldine, Raymond Wilson and Paul Smolensky (1993) An Optimality-Theoretic typology of case and grammatical voice systems. Proceedings of the 19th Annual Meeting of the Berkeley Linguistics Society pp464-478 Li, Charles (ed. 1976) Subject and Topic New York: Academic Press Li, Paul Jen-kuei (1973) Rukai Structure Taipei: Academia Sinica (Academia Sinica special publications 64) Lopez, Cecilio (1941) A manual of the Philippine national language (3rd ed) Manila: Bureau of Printing Lyons, John (1977) Semantics Cambridge UK: Cambridge University Press Maclachlan, Anna & Masanori Nakamura (1997) Case-checking and specificity in Tagalog The Linguistic Review 14: 307-333 Mahdi, Waruno (1988) Morphophonologische Besonderheiten und Historische Phonologie des Malagasy Berlin: Dietrich Reimer Verlag Mahdi, Waruno (1998) Posting to Bahasa list, 13 February 1998. Available from: http://w3.rz-berlin.mpg.de/~wm/PAP/BHS-L/BHS-adjectives.html Manning, Christopher D. (1996a) Ergativity: Argument Strcuture and Grammatical Relations Stanford, CA: CSLI Publications Manning, Christopher D. (1996b) Argument structure as the locus for binding theory. Presented at LFG Colloquium, Grenoble Manning, Christopher D. & Ivan A. Sag (1999) Dissociations between Argument Structure and Grammatical Relations. In Gert Webelhuth, Jean-Pierre Koenig and Andreas Kathol (eds.), Lexical And Constructional Aspects of Linguistic Explanation, CSLI Publications, pp. 63-78 Martin, J.R. (1996) Transitivity in Tagalog: A Functional Interpretation of Case. In Berry, Butler, Fawcett and Huang (ed.) Form and Meaning pp229-296, Norwood NJ: Ablex Publishing Corporation McCrae, Ken, Todd R. Ferretti & Liane Amyote (1997) Thematic roles as verbspecific concepts Language and Cognitive Processes 12: 137-176 248 REFERENCES McCune, Keith (1979) Passive Function and the Indonesian Passive. Oceanic Linguistics 18:119-169 McDonald, Ross D. & Soenjono Darjowidjojo (1967) Indonesian Reference Grammar Washington DC: Georgetown University Press Mel'cuk, Igor (1983) Grammatical subject and the problem of the ergative constructions in Lezgian Folio Slavica 5: 246-293 Miller, P. (1992) Postlexical cliticization v. affixation: co-ordination criteria. Papers from the 30th Chicago Linguistic Society 30: 382-396. Mithun, Marianne (1984) The Evolution of Noun Incorporation. Language 60:847-894 Moeliono, Anton M. (1994) Contact-induced language change in present-day Indonesian. In Tom Dutton & Darrell T.Tryon (ed) Language Contact and Change in the Austronesian World pp377-388, Berlin/New York: Mouton de Gruyter Moeliono, Anton M. & S. Dardjowidjojo (ed. 1988) Tata Bahasa Baku Bahasa Indonesia (Standard Grammar of Bahasa Indonesia) Jakarta: Balai Pustaka & Departemen Pendidikan dan Kebudayaan Moro, Andrea (1997) The raising of predicates Cambridge UK: Cambridge University Press Musgrave, Simon (2000) A Note on Animacy Hierarchy Effects in Sasak and Sumbawan. In P. Austin (ed) Working Papers in Sasak Volume 2 pp49-56 Myhill, John (1988) Nominal agent incorporation in Indonesian. Journal of Linguistics 24:111-136 Nakamura, Masanori (1998) Reference Set, Minimal Link Condition, and Parametrization. In P.Barbosa, D.Fox, P.Hagstrom, M.McGinnis and D.Pesetsky (eds) Is the best good enough? Optimality and Competition in Syntax Cambridge MA: MIT Press Naylor, Paz B. (1980) Linking, Relation Marking, and Tagalog Syntax. In P.B.Naylor (ed) Austronesian Syntax (Papers from the 2nd Eastern Conference on Austronesian Languages, Ann Arbor 1976) Ann Arbor: Center for South and Southeast Asian Studies (Michigan Papers on South and Southeast Asia, No.15) Naylor, Paz B. (1995) Subject, Topic and Tagalog Syntax. In David C. Bennett, Theodora Bynon and G. George Hewitt (eds) Subject, Voice and Ergativity: Selected Essays pp161-201, London: School of Oriental and African Studies Payne, Doris L. & Immanuel Barshi (1999) External possession: what, where, how and why. In Doris L. Payne & Immanuel Barshi (ed) External Possession pp3-29, Amsterdam/Philadelphia: John Benjamins Payne, Thomas E. (1982) Role and Reference related subject properties and ergativity in Yup'ik Eskimo and Tagalog Studies in Language 6: 75-100 Perlmutter, David & Paul Postal (1983) Some basic laws of clause structure. In David Perlmutter (ed) Studies in Relational Grammar I pp81-128, Chicago: University of Chicago Press Pesetsky, David (1995) Zero Syntax Cambridge MA: MIT Press Plank, Frans (ed. 1984) Objects: towards a theory of grammatical relations New York: Academic Press 249 REFERENCES Poedjosoedarmo, Gloria (1986) Subject selection and subject shifting in Indonesian. In Dardjowidjojo (ed) Miscellaneous Studies of Indonesian and Other Languages of Indonesia, Part VIII (NUSA 26) pp1-17 Pollock, Jean-Yves (1989) Verb movement, Universal Grammar and the structure of IP Linguistic Inquiry 20: 365-424 Poser, William J. (1992) Blocking of phrasal constructions by lexical items. In Ivan A. Sag and Anna Szabolsci (ed) Lexical Matters pp111-130, Stanford CA: CSLI Publications Postal, Paul M. (1969) On So-Called Pronouns in English, In D.Reibel & S.Schane (ed.) Modern Studies in English Eaglewood Cliffs NJ: Prentice-Hall pp201–224. Postal, Paul M. (1977) Antipassive in French Lingvisticae Investigationaes 1: 333-374 Prince, Alan & Paul Smolensky (1993) Optimality Theory: Constraint interaction in generative grammar RuCCs Technical Report #2. Rutgers University Center for Cognitive Science Pustet, Regina (2000) How arbitrary is lexical categorization? Verbs v. adjectives Linguistic Typology 4:175-212 Rosen, Carol G. (1984) The interface between semantic roles and initial grammatical relations. In David M. Perlmutter & Carol G. Rosen (ed) Studies in Relational Grammar 2 pp38-77. Chicago: The University of Chicago Press Sadler, Louisa (1998) Syntactic clitics, the structure-function mapping and morphological blocking. MS, University of Essex Schachter, Paul (1976) The subject in Philippine languages: Actor, Tpoic, Actor-Topic, or none of the above?. In Charles Li (ed.) Subject and Topic pp493–518. New York: Academic Press Schachter, Paul (1995) The subject in Tagalog: still none of the above. UCLA Occasional Papers in Linguistics No. 15 Los Angeles: Department of Linguistics, UCLA Schachter, Paul & Fe T. Otanes (1972) Tagalog Reference Grammar Berkeley: University of California Press Seiter, William J. (1978) On the syntactic character of middle objects in Polynesian. In S.A.Wurm and Lois Carrington (ed) Second International Conference on Austronesian Linguistics: Proceedings, Fascicle 2 pp 12891306, Canberra: Australian National University (Pacific Linguistics C-61) Sells, Peter (1998) The functions of voice markers in the Philippine languages. In S.G.Lapointe, D.K Brentari & P.M. Farrell (ed) Morphology and Its Relation to Phonology and Syntax pp111-137, Stanford CA: CSLI Publications Sells, Peter (to appear) Form and function in the typology of grammatical voice systems. In J.Grimshaw, G. Legendre & S.Vikner (ed) Optimality-theoretic Syntax Cambridge MA: MIT Press Seuren, Pieter A.M. (1998) Western Linguistics: An Historical Introduction Oxford: Blackwell Publishers Shibatani, Masayoshi (1985) Passives and related constructions: A prototype analysis. Language 61:821-848 250 REFERENCES Shibatani, Masayoshi (1994) An integrational approach to Possessor Raising, Ethical Datives, and Adversative Passives Proceedings of the Annual Meeting of the Berkeley Linguistic Society 20: 461-486 Silverstein, Michael (1976) Hierarchy of features and ergativity. In R.M.W. Dixon (ed) Grammatical Categories in Australian Languages pp112-171, Canberra: Australian Institute of Aboriginal Studies Simatupang, Iwan (1993) Merah merahnya Jakarta: Haji Masagung Smith, Henry (1996) Restrictiveness in Case Theory Cambridge UK: Cambridge University Press Sneddon, James Neil (1996) Indonesian Reference Grammar St Leonards NSW: Allen & Unwin Sneddon, James Neil (2000) Another look at ke- -an verbs. In Bambang Kaswanti Purwo (ed) Kajian Serba Linguistik untuk Anton Moeliono: Pereksa Bahasa pp511- 529, Jakarta/Atma Jaya: Universitas Katolik Indonesia and PT BPK Gunung Media Soemarmo (1970) Subject-Predicate, Focus-Presupposition and Topic-Comment in Bahasa Indonesia and Javanese PhD dissertation, University of California, Los Angeles Sproat, Richard (1985) Welsh syntax and VSO structure Natural Language and Linguistic Theory 2:173-216 Starosta, Stanley, Andrew K. Pawley & Lawrence A. Reid (1982) The evolution of focus in Austronesian. In Amran Halim, Lois Carrington & S.A.Wurm (ed) Papers from the Third International Conference on Austronesian Linguistics vol. 2 (Pacific Linguistics C-75) pp145-170, Canberra: Research School of Asian and Pacific Studies Steinhauer, Hein (1994) The Indonesian language situation and linguistics Bijdragen tot de Taal-, Land- en Volkenkunde 150: 755-784 Stevens, Alan M. (1970) Pseudo-transitive verbs in Indonesian. Indonesia 9:6772 Stowell, Tim (1983) Subjects across categories The Linguistic Review 2: 285312 Sukamto, Katarina (1999) The accessibility of inferrables in Indonesian. Austronesian Informal Seminar Series, The University of Melbourne, 30 April 1999. Tampubolon, Daulat P. (1983) Verbal Affixation in Indonesian, a semantic exploration (Pacific Linguistics D-48) Canberra: Research School of Pacific and Asian Studies Tanner, N. (1967) Speech and society among the Indonesian élite: a case study of a multilingual community Anthropological Linguistics 9: 15-40 Teeuw, A. (1962) Some problems in the study of word-classes in Bahasa Indonesia Lingua 9: 409-421 Teeuw, A. (1977) The morphological system of the Indonesian adjective. In Amran Halim (ed) Miscellaneous Studies in Indonesian and Languages in Indonesia Part II (NUSA 3) pp1-18 Tenny, Carol (1994) Aspectual Roles and the Syntax-Semantics Interface Dordrecht: Kluwer 251 REFERENCES Thomas, Michael R. (1978) Indonesian's unmarked verbs. In John M. Verhaar (ed) Miscellaneous Studies in Indonesian and Languages in Indonesia Part V (NUSA 6), pp 7-10 Thompson, Sandra (1988) A discourse approach to the cross-linguistic category 'adjective'. In John A. Hawkins (ed) Explaining Language Universals pp167-185, Oxford: Blackwells Tryon, Darrell (1995) Proto-Austronesian and the major Austronesian subgroups. In Peter Bellwood, James J. Fox and Darrell Tryon (ed) The Austronesians: Historical and Comparative Perspectives Canberra: The Australian Nationalk University, pp17-38 van den Berg, René (1996) The demise of focus and the spread of conjugated verbs. In H. Steinhauer (ed) Papers in Austronesian Linguistics No. 3 pp89114, Canberra: ANU (Pacific Linguistics A-84). van der Tuuk, H.N. (1971) A Grammar of Toba Batak The Hague: Martinus Nijhoff (Translation by J. Scott-Kemball of Tobaasche Spraakkunst Amsterdam 1864 and 1867) Van Valin, Robert D. Jr. & Randy LaPolla (1997) Syntax: Structure, meaning and function Cambridge UK: Cambridge University Press Verhaar, John W.M. (1984a) Affixation in Contemporary Indonesian. In Bambang Kaswanti Purwo (ed.) Towards a Description of Contemporary Indonesian: Preliminary Studies Part 1 (NUSA 18) pp1–26 Verhaar, John W.M. (1984b) The Categorial System in Contemporary Indonesian: Verbs. In Bambang Kaswanti Purwo (ed.) Towards a Description of Contemporary Indonesian: Preliminary Studies Part 1 (NUSA 18) pp27–63 Voskuil, Jan (1996) Comparative Morphology: Verb Taxonomy in Indonesian, Tagalog and Dutch The Hague: Holland Academic Graphics (HIL Dissertations No 21). Wechsler, Stephen, and I Wayan Arka (1998) Syntactic Ergativity in Balinese: an Argument Structure Based Theory. Natural Language and Linguistic Theory 16:387-441 Wierzbicka, Anna (1995) Adjectives vs. verbs: the iconicity of part-of-speech membership. In M. E. Landsberg (ed) Syntactic iconicity and linguistic freezes, pp223-245. Berlin/New York: Mouton de Gruyter. Winstedt, R.O. (1913) Malay Grammar Oxford: Clarendon Press Zaenen, Annie, Joan Maling & Hoskildur Thráinsson (1985) Case and grammatical function: the Icelandic passive. Natural Language and Linguistics Theory 3: 441-483 Zoetmulder, P.J. (1982) Old Javanese - English Dictionary s'Gravenhage: M.Nijhof 252